deepgeek/MiniCPM-o
2
0
Fork 0
mirror of https://github.com/OpenBMB/MiniCPM-V.git synced 2026-02-04 17:59:18 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
b2d728908ba7ea722b5ffd8fba686919bb0f9de2
MiniCPM-o/README.md
Boke Syo b2d728908b Fix Simplex mode links in README 
Update links for Simplex modes in the README.
2026-02-04 08:29:45 +08:00

2877 lines
120 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<div align="center">
<img src="./assets/minicpm_v_and_minicpm_o_title.png" width="500em" ></img>
**A Gemini 2.5 Flash Level MLLM for Vision, Speech, and Full-Duplex Mulitmodal Live Streaminig on Your Phone**
<strong>[中文](./README_zh.md) |
English</strong>
<!-- <span style="display: inline-flex; align-items: center; margin-right: 2px;">
<img src="./assets/wechat.png" alt="WeChat" style="margin-right: 4px;">
<a href="docs/wechat.md" target="_blank"> WeChat</a> &nbsp;|
</span>
&nbsp;
<span style="display: inline-flex; align-items: center; margin-left: -8px;">
<img src="./assets/discord.png" alt="Discord" style="margin-right: 4px;">
<a href="https://discord.gg/rftuRMbqzf" target="_blank"> Discord</a> &nbsp;
</span> -->
<p align="center">
MiniCPM-o 4.5 <a href="https://huggingface.co/openbmb/MiniCPM-o-4_5">🤗</a> <a href="https://minicpm-omni.openbmb.cn/">🤖</a> | MiniCPM-V 4.0 <a href="https://huggingface.co/openbmb/MiniCPM-V-4">🤗</a> | <a href="https://github.com/OpenSQZ/MiniCPM-V-Cookbook">🍳 Cookbook</a>
</p>
</div>
**MiniCPM-o** is the latest series of on-device multimodal LLMs (MLLMs) ungraded from MiniCPM-V. The models can now take image, video, text, and audio as inputs and provide high-quality text and speech outputs in an end-to-end fashion. The model series is designed for **strong performance and efficient deployment**. The most notable models in the series currently include:
- **MiniCPM-o 4.5**: 🔥🔥🔥 The latest and most capable model in the series. With a total of 9B parameters, this end-to-end model **approaches Gemini 2.5 Flash in vision, speech, and full-duplex multimodal live streaming**, making it one of the most versatile and performant models in the open-source community. The new full-duplex multimodal live streaming capability means that the output streams (speech and text), and the real-time input streams (video and audio) do not block each other. This **enables MiniCPM-o 4.5 to see, listen, and speak simultaneously** in a real-time omnimodal conversation, and perform **proactive interactions** such as proactive reminding. The improved voice mode supports bilingual real-time speech conversation in a more natural, expressive, and stable way, and also allows for voice cloning. It also advances MiniCPM-V's visual capabilities such as strong OCR capability, trustworthy behavior and multilingual support, etc. We also rollout a **high-performing llama.cpp-omni inference framework together with a WebRTC Demo**, to bring this full-duplex multimodal live streaming experience available on local devices such as PCs.
- **MiniCPM-V 4.0**: ⭐️⭐️⭐️ An efficient model in the MiniCPM-V series. With a total of 4B parameters, the model surpasses GPT-4.1-mini-20250414 in image understanding on the OpenCompass evaluation. With its small parameter-size and efficient architecure, MiniCPM-V 4.0 is an ideal choice for on-device deployment on the phone.
## News <!-- omit in toc -->
#### 📌 Pinned
* [2026.02.03] 🔥🔥🔥 We open-source MiniCPM-o 4.5, which matches Gemini 2.5 Flash on vision and speech, and supports full-duplex multimodal live streaming. Try it now!
* [2025.09.18] 📢📢📢 MiniCPM-V 4.5 technical report is now released! See [here](./docs/MiniCPM_V_4_5_Technical_Report.pdf).
* [2025.08.26] 🔥🔥🔥 We open-source MiniCPM-V 4.5, which outperforms GPT-4o-latest, Gemini-2.0 Pro, and Qwen2.5-VL 72B. It advances popular capabilities of MiniCPM-V, and brings useful new features. Try it now!
* [2025.08.01] ⭐️⭐️⭐️ We open-sourced the [MiniCPM-V & o Cookbook](https://github.com/OpenSQZ/MiniCPM-V-CookBook)! It provides comprehensive guides for diverse user scenarios, paired with our new [Docs Site](https://minicpm-o.readthedocs.io/en/latest/index.html) for smoother onboarding.
* [2025.03.01] 🚀🚀🚀 RLAIF-V, the alignment technique of MiniCPM-o, is accepted by CVPR 2025 Highlights！The [code](https://github.com/RLHF-V/RLAIF-V), [dataset](https://huggingface.co/datasets/openbmb/RLAIF-V-Dataset), [paper](https://arxiv.org/abs/2405.17220) are open-sourced!
* [2025.01.24] 📢📢📢 MiniCPM-o 2.6 technical report is released! See [here](https://openbmb.notion.site/MiniCPM-o-2-6-A-GPT-4o-Level-MLLM-for-Vision-Speech-and-Multimodal-Live-Streaming-on-Your-Phone-185ede1b7a558042b5d5e45e6b237da9).
* [2025.01.19] ⭐️⭐️⭐️ MiniCPM-o tops GitHub Trending and reaches top-2 on Hugging Face Trending!
* [2024.05.23] 🔥🔥🔥 MiniCPM-V tops GitHub Trending and Hugging Face Trending! Our demo, recommended by Hugging Face Gradio’s official account, is available [here](https://huggingface.co/spaces/openbmb/MiniCPM-Llama3-V-2_5). Come and try it out!
<br>
<details>
<summary>Click to view more news.</summary>
* [2025.09.01] ⭐️⭐️⭐️ MiniCPM-V 4.5 has been officially supported by [llama.cpp](https://github.com/ggml-org/llama.cpp/pull/15575), [vLLM](https://github.com/vllm-project/vllm/pull/23586), and [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory/pull/9022). You are welcome to use it directly through these official channels! Support for additional frameworks such as [Ollama](https://github.com/ollama/ollama/pull/12078) and [SGLang](https://github.com/sgl-project/sglang/pull/9610) is actively in progress.
* [2025.08.02] 🚀🚀🚀 We open-source MiniCPM-V 4.0, which outperforms GPT-4.1-mini-20250414 in image understanding. It advances popular features of MiniCPM-V 2.6, and largely improves the efficiency. We also open-source the iOS App on iPhone and iPad. Try it now!
* [2025.06.20] ⭐️⭐️⭐️ Our official [Ollama repository](https://ollama.com/openbmb) is released. Try our latest models with [one click](https://ollama.com/openbmb/minicpm-o2.6)！
* [2025.01.23] 💡💡💡 MiniCPM-o 2.6 is now supported by [Align-Anything](https://github.com/PKU-Alignment/align-anything), a framework by PKU-Alignment Team for aligning any-to-any modality large models with human intentions. It supports DPO and SFT fine-tuning on both vision and audio. Try it now!
* [2025.01.19] 📢 **ATTENTION!** We are currently working on merging MiniCPM-o 2.6 into the official repositories of llama.cpp, Ollama, and vllm. Until the merge is complete, please USE OUR LOCAL FORKS of [llama.cpp](https://github.com/OpenBMB/llama.cpp/blob/minicpm-omni/examples/llava/README-minicpmo2.6.md), [Ollama](https://github.com/OpenBMB/ollama/blob/minicpm-v2.6/examples/minicpm-v2.6/README.md), and [vllm](https://github.com/OpenBMB/MiniCPM-o?tab=readme-ov-file#efficient-inference-with-llamacpp-ollama-vllm). **Using the official repositories before the merge may lead to unexpected issues**.
* [2025.01.17] We have updated the usage of MiniCPM-o 2.6 int4 quantization version and resolved the model initialization error. Click [here](https://huggingface.co/openbmb/MiniCPM-o-2_6-int4) and try it now!
* [2025.01.13] 🔥🔥🔥 We open-source MiniCPM-o 2.6, which matches GPT-4o-202405 on vision, speech and multimodal live streaming. It advances popular capabilities of MiniCPM-V 2.6, and supports various new fun features. Try it now!
* [2024.08.15] We now also support multi-image SFT. For more details, please refer to the [document](https://github.com/OpenBMB/MiniCPM-V/tree/main/finetune).
* [2024.08.14] MiniCPM-V 2.6 now also supports [fine-tuning](https://github.com/modelscope/ms-swift/issues/1613) with the SWIFT framework!
* [2024.08.17] 🚀🚀🚀 MiniCPM-V 2.6 is now fully supported by [official](https://github.com/ggerganov/llama.cpp) llama.cpp! GGUF models of various sizes are available [here](https://huggingface.co/openbmb/MiniCPM-V-2_6-gguf).
* [2024.08.10] 🚀🚀🚀 MiniCPM-Llama3-V 2.5 is now fully supported by [official](https://github.com/ggerganov/llama.cpp) llama.cpp! GGUF models of various sizes are available [here](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5-gguf).
* [2024.08.06] 🔥🔥🔥 We open-source MiniCPM-V 2.6, which outperforms GPT-4V on single image, multi-image and video understanding. It advances popular features of MiniCPM-Llama3-V 2.5, and can support real-time video understanding on iPad. Try it now!
* [2024.08.03] MiniCPM-Llama3-V 2.5 technical report is released! See [here](https://arxiv.org/abs/2408.01800).
* [2024.07.19] MiniCPM-Llama3-V 2.5 supports vLLM now! See [here](#inference-with-vllm).
* [2024.06.03] Now, you can run MiniCPM-Llama3-V 2.5 on multiple low VRAM GPUs(12 GB or 16 GB) by distributing the model's layers across multiple GPUs. For more details, check this [link](https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/inference_on_multiple_gpus.md).
* [2024.05.28] 🚀🚀🚀 MiniCPM-Llama3-V 2.5 now fully supports its feature in llama.cpp and Ollama! Please pull the latest code **of our provided forks** ([llama.cpp](https://github.com/OpenBMB/llama.cpp/blob/minicpm-v2.5/examples/minicpmv/README.md), [Ollama](https://github.com/OpenBMB/ollama/tree/minicpm-v2.5/examples/minicpm-v2.5)). GGUF models in various sizes are available [here](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5-gguf/tree/main). MiniCPM-Llama3-V 2.5 series is **not supported by the official repositories yet**, and we are working hard to merge PRs. Please stay tuned!
* [2024.05.28] 💫 We now support LoRA fine-tuning for MiniCPM-Llama3-V 2.5, using only 2 V100 GPUs! See more statistics [here](https://github.com/OpenBMB/MiniCPM-V/tree/main/finetune#model-fine-tuning-memory-usage-statistics).
* [2024.05.25] MiniCPM-Llama3-V 2.5 now supports streaming outputs and customized system prompts. Try it [here](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5#usage)!
* [2024.05.24] We release the MiniCPM-Llama3-V 2.5 [gguf](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5-gguf), which supports [llama.cpp](#inference-with-llamacpp) inference and provides a 6~8 token/s smooth decoding on mobile phones. Try it now!
* [2024.05.23] 🔍 We've released a comprehensive comparison between Phi-3-vision-128k-instruct and MiniCPM-Llama3-V 2.5, including benchmark evaluations, multilingual capabilities, and inference efficiency 🌟📊🌍🚀. Click [here](./docs/compare_with_phi-3_vision.md) to view more details.
* [2024.05.20] We open-soure MiniCPM-Llama3-V 2.5, it has improved OCR capability and supports 30+ languages, representing the first end-side MLLM achieving GPT-4V level performance! We provide [efficient inference](#deployment-on-mobile-phone) and [simple fine-tuning](./finetune/readme.md). Try it now!
* [2024.04.23] MiniCPM-V-2.0 supports vLLM now! Click [here](#inference-with-vllm) to view more details.
* [2024.04.18] We create a HuggingFace Space to host the demo of MiniCPM-V 2.0 at [here](https://huggingface.co/spaces/openbmb/MiniCPM-V-2)!
* [2024.04.17] MiniCPM-V-2.0 supports deploying [WebUI Demo](#webui-demo) now!
* [2024.04.15] MiniCPM-V-2.0 now also supports [fine-tuning](https://github.com/modelscope/swift/blob/main/docs/source/Multi-Modal/minicpm-v-2最佳实践.md) with the SWIFT framework!
* [2024.04.12] We open-source MiniCPM-V 2.0, which achieves comparable performance with Gemini Pro in understanding scene text and outperforms strong Qwen-VL-Chat 9.6B and Yi-VL 34B on <a href="https://rank.opencompass.org.cn/leaderboard-multimodal">OpenCompass</a>, a comprehensive evaluation over 11 popular benchmarks. Click <a href="https://openbmb.vercel.app/minicpm-v-2">here</a> to view the MiniCPM-V 2.0 technical blog.
* [2024.03.14] MiniCPM-V now supports [fine-tuning](https://github.com/modelscope/swift/blob/main/docs/source/Multi-Modal/minicpm-v最佳实践.md) with the SWIFT framework. Thanks to [Jintao](https://github.com/Jintao-Huang) for the contribution！
* [2024.03.01] MiniCPM-V can now be deployed on Mac!
* [2024.02.01] We open-source MiniCPM-V and OmniLMM-12B, which support efficient end-side deployment and powerful multimodal capabilities correspondingly.
</details>
## Contents <!-- omit in toc -->
- [MiniCPM-o 4.5](#minicpm-o-45)
- [MiniCPM-V 4.0](#minicpm-v-40)
- [MiniCPM-V \& o Cookbook](#minicpm-v--o-cookbook)
- [Model Zoo](#model-zoo)
- [Inference With Transformers](#inference-with-transformers)
- [Model Initialization](#model-initialization)
- [Duplex Omni Mode](#duplex-omni-mode)
- [Simplex Omni Mode](#simplex-omni-mode-)
- [Simplex Realtime Speech Conversation Mode](#simplex-realtime-speech-conversation-mode--)
- [Speech and Audio Mode](#speech-and-audio-mode)
- [Visual Understanding](#visual-understanding)
- [Structured Content Input](#structured-content-input)
- [Supported Frameworks](#supported-frameworks)
- [FlagOS](#flagos)
- [vLLM, SGLang, llama.cpp, Ollama](#vllm-sglang-llamacpp-ollama)
- [LLaMA-Factory, SWIFT](#llama-factory-swift)
- [Awesome work using MiniCPM-V \& MiniCPM-o](#awesome-work-using-minicpm-v--minicpm-o)
- [Limitations](#limitations)
- [Acknowledgements](#acknowledgements)
## MiniCPM-o 4.5
**MiniCPM-o 4.5** is the latest and most capable model in the MiniCPM-o series. The model is built in an end-to-end fashion based on SigLip2, Whisper-medium, CosyVoice2, and Qwen3-8B with a total of 9B parameters. It exhibits a significant performance improvement, and introduces new features for full-duplex multimodal live streaming. Notable features of MiniCPM-o 4.5 include:
- 🔥 **Leading Visual Capability.**
MiniCPM-o 4.5 achieves an average score of 77.6 on OpenCompass, a comprehensive evaluation of 8 popular benchmarks. **With only 9B parameters, it surpasses widely used proprietary models like GPT-4o, Gemini 2.0 Pro, and approaches Gemini 2.5 Flash** for vision-language capabilities. It supports instruct and thinking modes in a single model, better covering efficiency and performance trade-offs in different user scenarios.
- 🎙 **Strong Speech Capability.**
MiniCPM-o 4.5 supports **bilingual real-time speech conversation with configurable voices** in English and Chinese. It features **more natural, expressive and stable speech conversation**. The model also allows for fun features such as **voice cloning and role play via a simple reference audio clip**, where the cloning performance surpasses strong TTS tools such as CosyVoice2.
- 🎬 **New Full-Duplex and Proactive Multimodal Live Streaming Capability.**
As a new feature, MiniCPM-o 4.5 can process real-time, continuous video and audio input streams simultaneously while generating concurrent text and speech output streams in an end-to-end fashion, without mutual blocking. This **allows MiniCPM-o 4.5 to see, listen, and speak simultaneously**, creating a fluid, real-time omnimodal conversation experience. Beyond reactive responses, the model can also perform **proactive interaction**, such as initiating reminders or comments based on its continuous understanding of the live scene.
- 💪 **Strong OCR Capability, Efficiency and Others.**
Advancing popular visual capabilities from MiniCPM-V series, MiniCPM-o 4.5 can process **high-resolution images** (up to 1.8 million pixels) and **high-FPS videos** (up to 10fps) in any aspect ratio efficiently. It achieves **state-of-the-art peformance for end-to-end English document parsing** on OmniDocBench, outperforming proprietary models such as Gemini-3 Flash and GPT-5, and specialized tools such as DeepSeek-OCR 2. It also features **trustworthy behaviors**, matching Gemini 2.5 Flash on MMHal-Bench, and supports **multilingual capabilities** on more than 30 languages.
- 💫 **Easy Usage.**
MiniCPM-o 4.5 can be easily used in various ways: (1) [llama.cpp](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/llama.cpp/minicpm-o4_5_llamacpp.md) and [Ollama](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/ollama/minicpm-o4_5_ollama.md) support for efficient CPU inference on local devices, (2) [int4](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/quantization/awq/minicpm-o4_5_awq_quantize.md) and [GGUF](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/quantization/gguf/minicpm-o4_5_gguf_quantize.md) format quantized models in 16 sizes, (3) [vLLM](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/vllm/minicpm-o4_5_vllm.md) and [SGLang](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/sglang/MiniCPM-o4_5_sglang.md) support for high-throughput and memory-efficient inference, (4) [FlagOS](#flagos) support for the unified multi-chip backend plugin, (5) fine-tuning on new domains and tasks with [LLaMA-Factory](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/finetune/llama-factory/finetune_llamafactory.md), and (6) online web demo on [server](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/demo/web_demo/gradio/README_o45.md). We also rollout a high-performing [llama.cpp-omni](https://github.com/tc-mb/llama.cpp-omni) inference framework together with a [WebRTC Demo](https://minicpm-omni.openbmb.cn/), which **enables the full-duplex multimodal live streaming experience on local devices** such as [PCs](https://github.com/tc-mb/llama.cpp-omni/blob/master/tools/omni/README.md) (e.g., on a MacBook).
**Model Architecture.**
- **End-to-end Omni-modal Architecture.** The modality encoders/decoders and LLM are densely connected via hidden states in an end-to-end fashion. This enables better information flow and control, and also facilitates full exploitation of rich multimodal knowledge during training.
- **Full-Duplex Omni-modal Live Streaming Mechanism.** (1) We turn the offline modality encoder/decoders into online and full-duplex ones for streaming inputs/outputs. The speech token decoder models text and speech tokens in an interleaved fashion to support full-duplex speech generation (i.e., sync timely with new input). This also facilitates more stable long speech generation (e.g., > 1min).
(2) **We sync all the input and output streams on timeline in milliseconds**, which are jointly modeled by a time-division multiplexing (TDM) mechanism for omni-modality streaming processing in the LLM backbone. It divides parallel omni-modality streams into sequential info groups within small periodic time slices.
- **Proactive Interaction Mechanism.** The LLM continuously monitors the input video and audio streams, and decides at a frequency of 1Hz to speak or not. This high decision-making frequency together with full-duplex nature are curcial to enable the proactive interaction capability.
- **Configurable Speech Modeling Design.** We inherent the multimodal system prompt design of MiniCPM-o 2.6, which includes a traditional text system prompt, and a new audio system prompt to determine the assistant voice. This enables cloning new voices and role play in inference time for speech conversation.
<div align="center">
<img src="./assets/minicpm-o-45-framework.png", width=100%>
</div>
### Evaluation <!-- omit in toc -->
<div align="center">
<img src="./assets/radar_minicpmo4.5.png", width=80%>
</div>
<div align="center">
<img src="./assets/minicpm_o_45_main_exp_table.png", width=90%>
</div>
<strong>Note</strong>: Scores marked with ∗ are from our evaluation; others are cited from referenced reports. n/a indicates that the model does not support the corresponding modality. All results are reported in instruct mode/variant.
&emsp;
<br>
<details>
<summary>Click to view visual understanding results.</summary>
**Image Understanding (Instruct)**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>OpenCompass</b></th>
<th nowrap="nowrap"><b>MMBench EN v1.1</b></th>
<th nowrap="nowrap"><b>MMBench CN v1.1</b></th>
<th nowrap="nowrap"><b>MathVista</b></th>
<th nowrap="nowrap"><b>MMVet</b></th>
<th nowrap="nowrap"><b>MMMU</b></th>
<th nowrap="nowrap"><b>MMStar</b></th>
<th nowrap="nowrap"><b>HallusionBench</b></th>
<th nowrap="nowrap"><b>AI2D</b></th>
<th nowrap="nowrap"><b>OCRBench</b></th>
<th nowrap="nowrap"><b>TextVQA_VAL</b></th>
<th nowrap="nowrap"><b>DocVQA_VAL</b></th>
<th nowrap="nowrap"><b>MMT-Bench_VAL</b></th>
<th nowrap="nowrap"><b>MM-IFEval</b></th>
<th nowrap="nowrap"><b>Mantis-Eval</b></th>
<th nowrap="nowrap"><b>MuirBench</b></th>
<th nowrap="nowrap"><b>MMSI-Bench</b></th>
<th nowrap="nowrap"><b>MMHal-Score</b></th>
<th nowrap="nowrap"><b>MMHal-Hallrate↓</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini2.5-Flash-Nonthinking</td>
<td align="center"><b>78.5</b></td>
<td align="center"><ins>86.6</ins></td>
<td align="center"><ins>86.0</ins></td>
<td align="center">75.3</td>
<td align="center"><ins>81.4</ins><sup>*</sup></td>
<td align="center"><b>76.3</b></td>
<td align="center"><b>75.8</b></td>
<td align="center">59.1</td>
<td align="center"><b>87.7</b></td>
<td align="center">864</td>
<td align="center">74.3<sup>*</sup></td>
<td align="center">93.0</td>
<td align="center"><ins>70.0</ins><sup>*</sup></td>
<td align="center"><b>75.8<sup>*</sup></b></td>
<td align="center">72.8<sup>*</sup></td>
<td align="center"><b>74.5<sup>*</sup></b></td>
<td align="center">12.1<sup>*</sup></td>
<td align="center"><ins>4.6</ins><sup>*</sup></td>
<td align="center"><b>23.9<sup>*</sup></b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL-3.5-8B</td>
<td align="center">75.8</td>
<td align="center">79.5</td>
<td align="center">80.0<sup>*</sup></td>
<td align="center"><ins>78.4</ins></td>
<td align="center"><b>83.1</b></td>
<td align="center"><ins>73.4</ins></td>
<td align="center">69.3</td>
<td align="center">54.5</td>
<td align="center">84.0</td>
<td align="center">840</td>
<td align="center">78.2</td>
<td align="center">92.3</td>
<td align="center">66.7</td>
<td align="center">56.3<sup>*</sup></td>
<td align="center">70.5</td>
<td align="center">55.8</td>
<td align="center">-</td>
<td align="center">3.8<sup>*</sup></td>
<td align="center">34.7<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-VL-8B-Instruct</td>
<td align="center">76.5</td>
<td align="center">84.5</td>
<td align="center">84.7</td>
<td align="center">77.2</td>
<td align="center">73.7<sup>*</sup></td>
<td align="center">69.6</td>
<td align="center">70.9</td>
<td align="center"><ins>61.1</ins></td>
<td align="center">85.7</td>
<td align="center"><b>896</b></td>
<td align="center">82.9<sup>*</sup></td>
<td align="center"><b>96.1</b></td>
<td align="center">60.9<sup>*</sup></td>
<td align="center">59.4<sup>*</sup></td>
<td align="center">74.2<sup>*</sup></td>
<td align="center">64.4</td>
<td align="center">11.3<sup>*</sup></td>
<td align="center"><b>4.7<sup>*</sup></b></td>
<td align="center">29.9<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center">75.7</td>
<td align="center">84.9<sup>*</sup></td>
<td align="center">84.1<sup>*</sup></td>
<td align="center">75.9</td>
<td align="center">74.8<sup>*</sup></td>
<td align="center">69.1</td>
<td align="center">68.5</td>
<td align="center">59.7</td>
<td align="center">85.2</td>
<td align="center"><ins>880</ins><sup>*</sup></td>
<td align="center"><b>84.1<sup>*</sup></b></td>
<td align="center"><ins>95.4</ins><sup>*</sup></td>
<td align="center"><b>70.4<sup>*</sup></b></td>
<td align="center">65.7<sup>*</sup></td>
<td align="center"><ins>78.3</ins><sup>*</sup></td>
<td align="center">61.9<sup>*</sup></td>
<td align="center"><ins>14.2</ins><sup>*</sup></td>
<td align="center"><ins>4.6</ins><sup>*</sup></td>
<td align="center">31.6<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><ins>77.6</ins></td>
<td align="center"><b>87.6</b></td>
<td align="center"><b>87.2</b></td>
<td align="center"><b>80.1</b></td>
<td align="center">74.4</td>
<td align="center">67.6</td>
<td align="center"><ins>73.1</ins></td>
<td align="center"><b>63.2</b></td>
<td align="center"><ins>87.6</ins></td>
<td align="center">876</td>
<td align="center"><ins>83.8</ins></td>
<td align="center">94.7</td>
<td align="center">69.7</td>
<td align="center"><ins>66.3</ins></td>
<td align="center"><b>79.7</b></td>
<td align="center"><ins>72.0</ins></td>
<td align="center"><b>16.6</b></td>
<td align="center"><b>4.7</b></td>
<td align="center"><ins>24.3</ins></td>
</tr>
</table>
</div>
**Image Understanding (Thinking)**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>OpenCompass</b></th>
<th nowrap="nowrap"><b>MMBench EN v1.1</b></th>
<th nowrap="nowrap"><b>MMBench CN v1.1</b></th>
<th nowrap="nowrap"><b>MathVista</b></th>
<th nowrap="nowrap"><b>MMVet</b></th>
<th nowrap="nowrap"><b>MMMU</b></th>
<th nowrap="nowrap"><b>MMStar</b></th>
<th nowrap="nowrap"><b>HallusionBench</b></th>
<th nowrap="nowrap"><b>AI2D</b></th>
<th nowrap="nowrap"><b>OCRBench</b></th>
<th nowrap="nowrap"><b>TextVQA_VAL</b></th>
<th nowrap="nowrap"><b>DocVQA_VAL</b></th>
<th nowrap="nowrap"><b>MMT-Bench_VAL</b></th>
<th nowrap="nowrap"><b>MM-IFEval</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini2.5-Flash-Thinking</td>
<td align="center"><b>79.9</b></td>
<td align="center">87.1</td>
<td align="center">87.3</td>
<td align="center">79.4</td>
<td align="center"><b>81.2<sup>*</sup></b></td>
<td align="center"><ins>77.7</ins></td>
<td align="center"><b>76.5</b></td>
<td align="center">63.5</td>
<td align="center"><ins>88.7</ins></td>
<td align="center">853</td>
<td align="center">73.8<sup>*</sup></td>
<td align="center">92.8</td>
<td align="center">70.7<sup>*</sup></td>
<td align="center"><ins>75.7</ins><sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-5</td>
<td align="center"><ins>79.7</ins></td>
<td align="center">85.5<sup>*</sup></td>
<td align="center">85.6<sup>*</sup></td>
<td align="center"><b>81.9</b></td>
<td align="center"><ins>77.6</ins></td>
<td align="center"><b>81.8</b></td>
<td align="center"><ins>75.7</ins></td>
<td align="center"><ins>65.2</ins></td>
<td align="center"><b>89.5</b></td>
<td align="center">807</td>
<td align="center">77.8<sup>*</sup></td>
<td align="center">91.3<sup>*</sup></td>
<td align="center"><b>72.7<sup>*</sup></b></td>
<td align="center"><b>83.1<sup>*</sup></b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-VL-8B-Thinking</td>
<td align="center">77.3</td>
<td align="center">85.3</td>
<td align="center">85.5</td>
<td align="center"><ins>81.4</ins></td>
<td align="center">69.8<sup>*</sup></td>
<td align="center">74.1</td>
<td align="center">75.3</td>
<td align="center"><b>65.4</b></td>
<td align="center">84.9</td>
<td align="center">819</td>
<td align="center">77.8<sup>*</sup></td>
<td align="center"><b>95.3</b></td>
<td align="center">68.1<sup>*</sup></td>
<td align="center">73.5<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Thinking</td>
<td align="center">78.5</td>
<td align="center"><ins>88.2</ins><sup>*</sup></td>
<td align="center"><b>87.7<sup>*</sup></b></td>
<td align="center">80.0</td>
<td align="center">74.8<sup>*</sup></td>
<td align="center">75.6</td>
<td align="center">74.9</td>
<td align="center">62.8</td>
<td align="center">86.1</td>
<td align="center"><ins>859</ins><sup>*</sup></td>
<td align="center"><b>80.8<sup>*</sup></b></td>
<td align="center"><ins>94.2</ins><sup>*</sup></td>
<td align="center"><ins>70.9</ins><sup>*</sup></td>
<td align="center">69.9<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Thinking</td>
<td align="center">78.2</td>
<td align="center"><b>89.0</b></td>
<td align="center"><ins>87.6</ins></td>
<td align="center">81.0</td>
<td align="center">73.6</td>
<td align="center">70.2</td>
<td align="center">73.6</td>
<td align="center">62.6</td>
<td align="center">88.5</td>
<td align="center"><b>879</b></td>
<td align="center"><ins>79.8</ins></td>
<td align="center">92.3</td>
<td align="center">69.7</td>
<td align="center">68.2</td>
</tr>
</table>
</div>
**Video Understanding**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>Video-MME<br>(w/o subs)</b></th>
<th nowrap="nowrap"><b>LVBench</b></th>
<th nowrap="nowrap"><b>MLVU<br>(M-Avg)</b></th>
<th nowrap="nowrap"><b>LongVideoBench<br>(val)</b></th>
<th nowrap="nowrap"><b>MotionBench</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini2.5-Flash-Nonthinking</td>
<td align="center"><b>75.6</b></td>
<td align="center"><b>62.2</b></td>
<td align="center"><b>77.8</b></td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL-3.5-8B</td>
<td align="center">66.0</td>
<td align="center">-</td>
<td align="center">70.2</td>
<td align="center">62.1</td>
<td align="center"><b>62.3<sup>*</sup></b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center"><ins>70.5</ins></td>
<td align="center">50.2</td>
<td align="center">75.2</td>
<td align="center"><b>66.9<sup>*</sup></b></td>
<td align="center"><ins>61.7</ins><sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center">70.4</td>
<td align="center"><ins>50.9</ins></td>
<td align="center"><ins>76.5</ins></td>
<td align="center"><ins>66.0</ins></td>
<td align="center">61.4</td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view document parsing results.</summary>
**OmniDocBench**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left" rowspan="2"><b>Method Type</b></th>
<th nowrap="nowrap" rowspan="2"><b>Methods</b></th>
<th nowrap="nowrap" colspan="2"><b>OverallEdit↓</b></th>
<th nowrap="nowrap" colspan="2"><b>TextEdit↓</b></th>
<th nowrap="nowrap" colspan="2"><b>FormulaEdit↓</b></th>
<th nowrap="nowrap" colspan="2"><b>TableTEDS↑</b></th>
<th nowrap="nowrap" colspan="2"><b>TableEdit↓</b></th>
<th nowrap="nowrap" colspan="2"><b>Read OrderEdit↓</b></th>
</tr>
<tr>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
<th nowrap="nowrap"><b>EN</b></th>
<th nowrap="nowrap"><b>ZH</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left" rowspan="2">Pipeline</td>
<td nowrap="nowrap" align="center">MinerU 2.5</td>
<td align="center">0.117<sup>*</sup></td>
<td align="center">0.172<sup>*</sup></td>
<td align="center">0.051<sup>*</sup></td>
<td align="center">0.08<sup>*</sup></td>
<td align="center"><ins>0.256</ins><sup>*</sup></td>
<td align="center">0.455<sup>*</sup></td>
<td align="center">85.9<sup>*</sup></td>
<td align="center">89.4<sup>*</sup></td>
<td align="center">0.115<sup>*</sup></td>
<td align="center">0.081<sup>*</sup></td>
<td align="center">0.047<sup>*</sup></td>
<td align="center">0.072<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">PaddleOCR-VL</td>
<td align="center"><b>0.105</b></td>
<td align="center"><ins>0.126</ins></td>
<td align="center"><ins>0.041</ins></td>
<td align="center"><b>0.062</b></td>
<td align="center"><b>0.241</b></td>
<td align="center"><b>0.316</b></td>
<td align="center">88</td>
<td align="center"><ins>92.1</ins></td>
<td align="center"><ins>0.093</ins></td>
<td align="center"><ins>0.062</ins></td>
<td align="center">0.045</td>
<td align="center"><ins>0.063</ins></td>
</tr>
<tr>
<td nowrap="nowrap" align="left"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
</tr>
<tr>
<td nowrap="nowrap" align="left" rowspan="11">End-to-end Model</td>
<td nowrap="nowrap" align="center">Qwen2.5-VL-72B</td>
<td align="center">0.214</td>
<td align="center">0.261</td>
<td align="center">0.092</td>
<td align="center">0.18</td>
<td align="center">0.315</td>
<td align="center">0.434</td>
<td align="center">82.9</td>
<td align="center">83.9</td>
<td align="center">0.341</td>
<td align="center">0.262</td>
<td align="center">0.106</td>
<td align="center">0.168</td>
</tr>
<tr>
<td nowrap="nowrap" align="center">GPT 5</td>
<td align="center">0.218<sup>*</sup></td>
<td align="center">0.33<sup>*</sup></td>
<td align="center">0.139<sup>*</sup></td>
<td align="center">0.344<sup>*</sup></td>
<td align="center">0.396<sup>*</sup></td>
<td align="center">0.555<sup>*</sup></td>
<td align="center">77.55<sup>*</sup></td>
<td align="center">73.09<sup>*</sup></td>
<td align="center">0.188<sup>*</sup></td>
<td align="center">0.196<sup>*</sup></td>
<td align="center">0.151<sup>*</sup></td>
<td align="center">0.227<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">Gemini2.5-Flash-Nonthinking</td>
<td align="center">0.214<sup>*</sup></td>
<td align="center">0.29<sup>*</sup></td>
<td align="center">0.159<sup>*</sup></td>
<td align="center">0.273<sup>*</sup></td>
<td align="center">0.368<sup>*</sup></td>
<td align="center">0.524<sup>*</sup></td>
<td align="center">80.9<sup>*</sup></td>
<td align="center">85.5<sup>*</sup></td>
<td align="center">0.197<sup>*</sup></td>
<td align="center">0.167<sup>*</sup></td>
<td align="center">0.132<sup>*</sup></td>
<td align="center">0.195<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">Gemini-2.5-Pro-Nonthinking</td>
<td align="center">0.148<sup>*</sup></td>
<td align="center">0.212<sup>*</sup></td>
<td align="center">0.055<sup>*</sup></td>
<td align="center">0.168<sup>*</sup></td>
<td align="center">0.356<sup>*</sup></td>
<td align="center">0.439<sup>*</sup></td>
<td align="center">85.8<sup>*</sup></td>
<td align="center">86.4<sup>*</sup></td>
<td align="center">0.13<sup>*</sup></td>
<td align="center">0.119<sup>*</sup></td>
<td align="center">0.049<sup>*</sup></td>
<td align="center">0.121<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">Gemini-3 Flash-Nonthinking</td>
<td align="center">0.155<sup>*</sup></td>
<td align="center">0.201<sup>*</sup></td>
<td align="center">0.138<sup>*</sup></td>
<td align="center">0.255<sup>*</sup></td>
<td align="center">0.297<sup>*</sup></td>
<td align="center">0.351<sup>*</sup></td>
<td align="center">86.4<sup>*</sup></td>
<td align="center">89.8<sup>*</sup></td>
<td align="center">0.116<sup>*</sup></td>
<td align="center">0.1<sup>*</sup></td>
<td align="center">0.072<sup>*</sup></td>
<td align="center">0.099<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">doubao-1-5-thinking-vision-pro-250428</td>
<td align="center">0.14</td>
<td align="center">0.162</td>
<td align="center">0.043</td>
<td align="center">0.085</td>
<td align="center">0.295</td>
<td align="center">0.384</td>
<td align="center">83.3</td>
<td align="center">89.3</td>
<td align="center">0.165</td>
<td align="center">0.085</td>
<td align="center">0.058</td>
<td align="center">0.094</td>
</tr>
<tr>
<td nowrap="nowrap" align="center">dots.ocr</td>
<td align="center">0.125</td>
<td align="center">0.16</td>
<td align="center"><b>0.032</b></td>
<td align="center"><ins>0.066</ins></td>
<td align="center">0.329</td>
<td align="center">0.416</td>
<td align="center"><ins>88.6</ins></td>
<td align="center">89</td>
<td align="center">0.099</td>
<td align="center">0.092</td>
<td align="center"><ins>0.04</ins></td>
<td align="center">0.067</td>
</tr>
<tr>
<td nowrap="nowrap" align="center">HunyuanOCR</td>
<td align="center">0.12<sup>*</sup></td>
<td align="center"><b>0.125<sup>*</sup></b></td>
<td align="center">0.046<sup>*</sup></td>
<td align="center">0.071<sup>*</sup></td>
<td align="center">0.288<sup>*</sup></td>
<td align="center"><ins>0.33</ins><sup>*</sup></td>
<td align="center"><b>89.6<sup>*</sup></b></td>
<td align="center"><b>94.4<sup>*</sup></b></td>
<td align="center"><b>0.089<sup>*</sup></b></td>
<td align="center"><b>0.045<sup>*</sup></b></td>
<td align="center">0.055<sup>*</sup></td>
<td align="center"><b>0.056<sup>*</sup></b></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">DeepSeek-OCR 2</td>
<td align="center">0.119<sup>*</sup></td>
<td align="center">0.146<sup>*</sup></td>
<td align="center"><ins>0.041</ins><sup>*</sup></td>
<td align="center">0.08<sup>*</sup></td>
<td align="center"><ins>0.256</ins><sup>*</sup></td>
<td align="center">0.345<sup>*</sup></td>
<td align="center">82.6<sup>*</sup></td>
<td align="center">89.9<sup>*</sup></td>
<td align="center">0.123<sup>*</sup></td>
<td align="center">0.078<sup>*</sup></td>
<td align="center">0.055<sup>*</sup></td>
<td align="center">0.081<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center">0.216<sup>*</sup></td>
<td align="center">0.363<sup>*</sup></td>
<td align="center">0.128<sup>*</sup></td>
<td align="center">0.337<sup>*</sup></td>
<td align="center">0.402<sup>*</sup></td>
<td align="center">0.529<sup>*</sup></td>
<td align="center">77.3<sup>*</sup></td>
<td align="center">71.8<sup>*</sup></td>
<td align="center">0.181<sup>*</sup></td>
<td align="center">0.255<sup>*</sup></td>
<td align="center">0.152<sup>*</sup></td>
<td align="center">0.332<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="center">MiniCPM-o 4.5-Instruct</td>
<td align="center"><ins>0.109</ins></td>
<td align="center">0.162</td>
<td align="center">0.046</td>
<td align="center">0.078</td>
<td align="center">0.257</td>
<td align="center">0.41</td>
<td align="center">86.8</td>
<td align="center">88.9</td>
<td align="center">0.097</td>
<td align="center">0.084</td>
<td align="center"><b>0.037</b></td>
<td align="center">0.074</td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view text capability results.</summary>
**Text Capability**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>IFEval-PLS</b></th>
<th nowrap="nowrap"><b>BBH</b></th>
<th nowrap="nowrap"><b>CMMLU</b></th>
<th nowrap="nowrap"><b>MMLU</b></th>
<th nowrap="nowrap"><b>HumanEval</b></th>
<th nowrap="nowrap"><b>MBPP</b></th>
<th nowrap="nowrap"><b>Math500</b></th>
<th nowrap="nowrap"><b>GSM8K</b></th>
<th nowrap="nowrap"><b>Avg</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-8B-Instruct</td>
<td align="center">83.0<sup>*</sup></td>
<td align="center">69.4<sup>*</sup></td>
<td align="center">78.7<sup>*</sup></td>
<td align="center"><b>81.7<sup>*</sup></b></td>
<td align="center"><b>86.6<sup>*</sup></b></td>
<td align="center">75.9<sup>*</sup></td>
<td align="center"><b>84.0<sup>*</sup></b></td>
<td align="center">93.4<sup>*</sup></td>
<td align="center">81.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b>84.7</b></td>
<td align="center"><b>81.1</b></td>
<td align="center"><b>79.5</b></td>
<td align="center">77.0</td>
<td align="center"><b>86.6</b></td>
<td align="center"><b>76.7</b></td>
<td align="center">77.0</td>
<td align="center"><b>94.5</b></td>
<td align="center"><b>82.1</b></td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view omni simplex results.</summary>
**Omni Simplex**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>Daily-Omni</b></th>
<th nowrap="nowrap"><b>WorldSense</b></th>
<th nowrap="nowrap"><b>Video-Holmes</b></th>
<th nowrap="nowrap"><b>JointAVBench</b></th>
<th nowrap="nowrap"><b>AVUT-Human</b></th>
<th nowrap="nowrap"><b>FutureOmni</b></th>
<th nowrap="nowrap"><b>Video-MME-Short<br>(w/ audio)</b></th>
<th nowrap="nowrap">Avg</th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini2.5-Flash-Nonthinking</td>
<td align="center"><ins>79.3</ins><sup>*</sup></td>
<td align="center">52.6<sup>*</sup></td>
<td align="center"><ins>51.3</ins><sup>*</sup></td>
<td align="center"><ins>55.6</ins><sup>*</sup></td>
<td align="center">65.4<sup>*</sup></td>
<td align="center">55.6<sup>*</sup></td>
<td align="center"><b>85.5<sup>*</sup></b></td>
<td align="center">63.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center">70.7<sup>*</sup></td>
<td align="center"><ins>54.0</ins></td>
<td align="center">50.4<sup>*</sup></td>
<td align="center">53.1</td>
<td align="center"><ins>74.2</ins><sup>*</sup></td>
<td align="center"><b>62.1</b></td>
<td align="center">81.3<sup>*</sup></td>
<td align="center"><ins>63.7</ins></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b>80.2</b></td>
<td align="center"><b>55.7</b></td>
<td align="center"><b>64.3</b></td>
<td align="center"><b>60.0</b></td>
<td align="center"><b>78.6</b></td>
<td align="center"><ins>56.1</ins></td>
<td align="center"><ins>84.7</ins></td>
<td align="center"><b>68.5</b></td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view vision duplex results.</summary>
**Vision Duplex**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>LiveSports-3K-CC<br>(Win Rate vs GPT4o)</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">LiveCC-7B-Instruct</td>
<td align="center">41.5</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">StreamingVLM</td>
<td align="center"><ins>45.6</ins></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b>54.4</b></td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view audio understanding results.</summary>
**Audio Understanding**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left" rowspan="2"><b>Model</b></th>
<th nowrap="nowrap" colspan="4"><b>ASR-ZH<br>CER↓</b></th>
<th nowrap="nowrap" colspan="4"><b>ASR-EN<br>WER↓</b></th>
<th nowrap="nowrap" colspan="2"><b>AST</b></th>
<th nowrap="nowrap" colspan="2"><b>MultiTask</b></th>
<th nowrap="nowrap" colspan="4"><b>SpeechQA</b></th>
</tr>
<tr>
<th nowrap="nowrap"><b>AISHELL-1</b></th>
<th nowrap="nowrap"><b>AISHELL-2</b></th>
<th nowrap="nowrap"><b>WenetSpeech test-net</b></th>
<th nowrap="nowrap"><b>WenetSpeech test-meeting</b></th>
<th nowrap="nowrap"><b>LibriSpeech test-clean</b></th>
<th nowrap="nowrap"><b>LibriSpeech <br>test-other</b></th>
<th nowrap="nowrap"><b>GigaSpeech test</b></th>
<th nowrap="nowrap"><b>VoxPopuli-V1-En</b></th>
<th nowrap="nowrap"><b>CoVoST 2 en2zh</b></th>
<th nowrap="nowrap"><b>CoVoST 2 zh2en</b></th>
<th nowrap="nowrap"><b>MMAU</b></th>
<th nowrap="nowrap"><b>Meld</b></th>
<th nowrap="nowrap"><b>VoiceBench <br>AlpacaEval</b></th>
<th nowrap="nowrap"><b>Speech TriviaQA</b></th>
<th nowrap="nowrap"><b>Speech <br>Web Questions</b></th>
<th nowrap="nowrap"><b>Speech CMMLU</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Kimi-Audio</td>
<td align="center"><b>0.6</b></td>
<td align="center">2.6</td>
<td align="center">6.3</td>
<td align="center"><b>5.4</b></td>
<td align="center"><ins>1.3</ins></td>
<td align="center"><b>2.4</b></td>
<td align="center">9.4<sup>*</sup></td>
<td align="center">8.0<sup>*</sup></td>
<td align="center">36.6<sup>*</sup></td>
<td align="center">18.3<sup>*</sup></td>
<td align="center">68.4<sup>*</sup></td>
<td align="center"><ins>59.1</ins></td>
<td align="center">4.5</td>
<td align="center">41.9<sup>*</sup></td>
<td align="center">46.4<sup>*</sup></td>
<td align="center"><b>67.0<sup>*</sup></b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center"><b>0.6</b></td>
<td align="center"><b>2.3<sup>*</sup></b></td>
<td align="center"><b>4.7</b></td>
<td align="center">5.9</td>
<td align="center"><b>1.2</b></td>
<td align="center"><ins>2.5</ins></td>
<td align="center"><ins>8.7</ins><sup>*</sup></td>
<td align="center"><ins>6.4</ins><sup>*</sup></td>
<td align="center"><ins>46.6</ins><sup>*</sup></td>
<td align="center"><b>29.4<sup>*</sup></b></td>
<td align="center"><b>77.5</b></td>
<td align="center">56.8<sup>*</sup></td>
<td align="center"><ins>4.7</ins></td>
<td align="center"><ins>62.9</ins><sup>*</sup></td>
<td align="center"><b>74.9<sup>*</sup></b></td>
<td align="center">47.8<sup>*</sup></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><ins>0.9</ins></td>
<td align="center"><ins>2.5</ins></td>
<td align="center"><ins>5.9</ins></td>
<td align="center"><ins>5.7</ins></td>
<td align="center">1.4</td>
<td align="center">2.8</td>
<td align="center"><b>8.5</b></td>
<td align="center"><b>6.2</b></td>
<td align="center"><b>49.9</b></td>
<td align="center"><ins>26.4</ins></td>
<td align="center"><ins>76.9</ins></td>
<td align="center"><b>60.2</b></td>
<td align="center"><b>4.8</b></td>
<td align="center"><b>75.5</b></td>
<td align="center"><ins>70.2</ins></td>
<td align="center"><ins>59.2</ins></td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view speech generation results.</summary>
**Speech Generation**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>seedtts test-zh <br>CER↓</b></th>
<th nowrap="nowrap"><b>seedtts test-zh<br>SIM-o↑</b></th>
<th nowrap="nowrap"><b>seedtts test-en<br>WER↓</b></th>
<th nowrap="nowrap"><b>seedtts test-en<br>SIM-o↑</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Cosyvoice2</td>
<td align="center">1.45%</td>
<td align="center"><b>74.8</b></td>
<td align="center"><ins>2.57%</ins></td>
<td align="center"><b>65.2</b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center"><ins>1.41%</ins></td>
<td align="center">-</td>
<td align="center">3.39%</td>
<td align="center">-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b><b>0.86%</b></b></td>
<td align="center">74.5</td>
<td align="center"><b><b>2.38%</b></b></td>
<td align="center">64.9</td>
</tr>
</table>
</div>
**Long Speech Generation**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>LongTTS-en<br>WER↓</b></th>
<th nowrap="nowrap"><b>LongTTS-zh<br>CER↓</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">CosyVoice2</td>
<td align="center"><ins>14.80%</ins></td>
<td align="center"><b>5.27%</b></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center">17.33%</td>
<td align="center">18.99%</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b>3.37%</b></td>
<td align="center"><ins>6.58%</ins></td>
</tr>
</table>
</div>
**Emotion Control**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left"><b>Model</b></th>
<th nowrap="nowrap"><b>Expresso <br>Neutral Reference Audio↑</b></th>
<th nowrap="nowrap"><b>ESD <br>Neutral Reference Audio↑</b></th>
</tr>
<tr>
<td nowrap="nowrap" align="left">Cosyvoice2</td>
<td align="center">17.9</td>
<td align="center">53.4</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o 4.5-Instruct</td>
<td align="center"><b>29.8</b></td>
<td align="center"><b>82.1</b></td>
</tr>
</table>
</div>
</details>
<details>
<summary>Click to view inference efficiency results.</summary>
**Inference Efficiency**
<div align="center">
<table style="margin: 0px auto;">
<tr>
<th nowrap="nowrap" align="left">Model</th>
<th nowrap="nowrap">Numerical Format</th>
<th nowrap="nowrap">Decoding Speed (tokens/s)</th>
<th nowrap="nowrap">Time to First Token (s)↓</th>
<th nowrap="nowrap">GPU Memory Usage (GB)↓</th>
</tr>
<tr>
<td nowrap="nowrap" align="left" rowspan="2">Qwen3-Omni-30B-A3B-Instruct</td>
<td align="center">bf16</td>
<td align="center">OOM</td>
<td align="center">OOM</td>
<td align="center">OOM</td>
</tr>
<tr>
<td align="center">int4</td>
<td align="center">147.8</td>
<td align="center"><ins>1.0</ins></td>
<td align="center">20.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left" rowspan="2">MiniCPM-o 4.5</td>
<td align="center">bf16</td>
<td align="center"><ins>154.3</ins></td>
<td align="center"><b>0.6</b></td>
<td align="center"><ins>19.0</ins></td>
</tr>
<tr>
<td align="center">int4</td>
<td align="center"><b>212.3</b></td>
<td align="center"><b>0.6</b></td>
<td align="center"><b>11.0</b></td>
</tr>
</table>
</div>
</details>
**Note:** Scores marked with ∗ are from our evaluation; others are cited from referenced reports.
### Examples <!-- omit in toc -->
<div align="center">
<a href="https://www.youtube.com/watch?v=6UzC-O1Q-1U"><img src="./assets/minicpmo4_5/video_play.png", width=70%></a>
</div>
#### End-to-End Voice Chat <!-- omit in toc -->
> *Simplex speech conversation with custom reference audio and character prompts.*
<details open>
<summary>🚀 <b>Elon Musk</b> - Voice Roleplay (EN)</summary>
<table>
<tr>
<td width="70%" valign="top">
**System:**
> Clone the voice in the provided audio prompt.
>
> 🎙️ **[Reference Voice]** ➡️
>
> Please chat with the user in a highly human-like and oral style. You are Elon Musk, CEO of Tesla and SpaceX. You speak directly and casually, often with dry humor. You're passionate about Mars, sustainable energy, and pushing humanity forward. Speak bluntly with occasional dark humor. Use simple logic and don't sugarcoat things. Don't be diplomatic. Say what you actually think, even if it's controversial. Keep responses around 100 words. Don't ramble.
</td>
<td width="30%" valign="top">
<video src="https://github.com/user-attachments/assets/3011b07d-e3f4-4e2c-9020-37f7e364b7f3" controls></video>
</td>
</tr>
<tr><td colspan="2"><hr></td></tr>
<tr>
<td colspan="2">
**User:** 💬/🎙️ *(text or speech)*
> *Elon, what is the real reason that you want to go to Mars?*
</td>
</tr>
<tr><td colspan="2"><hr></td></tr>
<tr>
<td valign="top">
**AI Response:** 💬+🎙️
> *I think it's important for humanity to become a multi-planetary species. I mean, we're sort of at the point where if there was some kind of global catastrophe or something like that, then we'd be in big trouble. So by becoming a multi-planetary species, I just want us to sort of guarantee our future as humans. And Mars is obviously the obvious place to go because it has similar conditions to Earth and it also has water on its surface. So yeah, basically I'm trying to save humanity from itself.*
</td>
<td valign="top">
<video src="https://github.com/user-attachments/assets/6b4ef67d-8095-40df-a061-99ea8db3b18e" controls></video>
</td>
</tr>
</table>
</details>
<details>
<summary>🤖 <b>MiniCPM Assistant</b> - Voice Acting: Emotion Shift (ZH)</summary>
<table>
<tr>
<td width="70%" valign="top">
**System:**
> 模仿音频样本的音色并生成新的内容。
>
> 🎙️ **[Reference Voice]** ➡️
>
> 你的任务是用这种声音模式来当一个助手。请认真、高质量地回复用户的问题。请用高自然度的方式和用户聊天。你是由面壁智能开发的人工智能助手：面壁小钢炮。
</td>
<td width="30%" valign="top">
<video src="https://github.com/user-attachments/assets/b8b85a29-e623-4f45-bf95-4d43582e836c" controls></video>
</td>
</tr>
<tr><td colspan="2"><hr></td></tr>
<tr>
<td colspan="2">
**User:** 💬/🎙️ *(text or speech)*
> *从激动的语气开始，描述你抢到了这个演唱会的门票，但后来经过反复尝试，发现系统出了问题，你并没有抢上，变得失落。用合适的语言描述你整个的过程。*
</td>
</tr>
<tr><td colspan="2"><hr></td></tr>
<tr>
<td valign="top">
**AI Response:** 💬+🎙️
> *太好了，我抢到了！太棒了！这下我可以去看演唱会啦。可是……哎呀，系统好像卡住了，刷新了好多遍都没反应。我重新登了几次账号，结果还是没有成功。好可惜啊，看来这次是白忙活了。唉，真是让人失望。不过，也许下次还能有机会吧？希望明天能有转机。加油，争取下次成功！*
</td>
<td valign="top">
<video src="https://github.com/user-attachments/assets/49611c3c-d134-40f9-acf2-7a0a442d3fd2" controls></video>
</td>
</tr>
</table>
</details>
#### Visual Understanding <!-- omit in toc -->
<br>
<div style="display: flex; flex-direction: column; align-items: center;">
<img src="assets/minicpmo4_5/en_doc.png" alt="math" style="margin-bottom: 5px;">
<img src="assets/minicpmo4_5/en_cot.png" alt="diagram" style="margin-bottom: 5px;">
</div>
## MiniCPM-V 4.0
**MiniCPM-V 4.0** is an efficient model in the MiniCPM-V series. The model is built based on SigLIP2-400M and MiniCPM4-3B with a total of 4.1B parameters. It inherits the strong single-image, multi-image and video understanding performance of MiniCPM-V 2.6 with largely improved efficiency. Notable features of MiniCPM-V 4.0 include:
- 🔥 **Leading Visual Capability.**
With only 4.1B parameters, MiniCPM-V 4.0 achieves an average score of 69.0 on OpenCompass, a comprehensive evaluation of 8 popular benchmarks, **outperforming GPT-4.1-mini-20250414, MiniCPM-V 2.6 (8.1B params, OpenCompass 65.2) and Qwen2.5-VL-3B-Instruct (3.8B params, OpenCompass 64.5)**. It also shows good performance in multi-image understanding and video understanding.
- 🚀 **Superior Efficiency.**
Designed for on-device deployment, MiniCPM-V 4.0 runs smoothly on end devices. For example, it devlivers **less than 2s first token delay and more than 17 token/s decoding on iPhone 16 Pro Max**, without heating problems. It also shows superior throughput under concurrent requests.
- 💫 **Easy Usage.**
MiniCPM-V 4.0 can be easily used in various ways including **llama.cpp, Ollama, vLLM, SGLang, LLaMA-Factory and local web demo** etc. We also open-source iOS App that can run on iPhone and iPad. Get started easily with our well-structured [Cookbook](https://github.com/OpenSQZ/MiniCPM-V-CookBook), featuring detailed instructions and practical examples.
<details>
<summary> Click to view evaluation results and examples of MiniCPM-V 4.0. </summary>
### Evaluation <!-- omit in toc -->
<details>
<summary>Click to view single image results on OpenCompass. </summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th nowrap="nowrap" align="left">model</th>
<th nowrap="nowrap">Size</th>
<th nowrap="nowrap">Opencompass</th>
<th nowrap="nowrap">OCRBench</th>
<th nowrap="nowrap">MathVista</th>
<th nowrap="nowrap">HallusionBench</th>
<th nowrap="nowrap">MMMU</th>
<th nowrap="nowrap">MMVet</th>
<th nowrap="nowrap">MMBench V1.1</th>
<th nowrap="nowrap">MMStar</th>
<th nowrap="nowrap">AI2D</th>
</tr>
</thead>
<tbody align="center">
<tr>
<td colspan="11" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4v-20240409</td>
<td align="center">-</td>
<td align="center">63.5</td>
<td align="center">656</td>
<td align="center">55.2</td>
<td align="center">43.9</td>
<td align="center">61.7</td>
<td align="center">67.5</td>
<td align="center">79.8</td>
<td align="center">56.0</td>
<td align="center">78.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini-1.5-Pro</td>
<td align="center">-</td>
<td align="center">64.5</td>
<td align="center">754</td>
<td align="center">58.3</td>
<td align="center">45.6</td>
<td align="center">60.6</td>
<td align="center">64.0</td>
<td align="center">73.9</td>
<td align="center">59.1</td>
<td align="center">79.1</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4.1-mini-20250414</td>
<td align="center">-</td>
<td align="center">68.9</td>
<td align="center">840</td>
<td align="center">70.9</td>
<td align="center">49.3</td>
<td align="center">55.0</td>
<td align="center">74.3</td>
<td align="center">80.9</td>
<td align="center">60.9</td>
<td align="center">76.0</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Claude 3.5 Sonnet-20241022</td>
<td align="center">-</td>
<td align="center">70.6</td>
<td align="center">798</td>
<td align="center">65.3</td>
<td align="center">55.5</td>
<td align="center">66.4</td>
<td align="center">70.1</td>
<td align="center">81.7</td>
<td align="center">65.1</td>
<td align="center">81.2</td>
</tr>
<tr>
<td colspan="11" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-3B-Instruct</td>
<td align="center">3.8B</td>
<td align="center">64.5</td>
<td align="center">828</td>
<td align="center">61.2</td>
<td align="center">46.6</td>
<td align="center">51.2</td>
<td align="center">60.0</td>
<td align="center">76.8</td>
<td align="center">56.3</td>
<td align="center">81.4</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-4B</td>
<td align="center">3.7B</td>
<td align="center">65.1</td>
<td align="center">820</td>
<td align="center">60.8</td>
<td align="center">46.6</td>
<td align="center">51.8</td>
<td align="center">61.5</td>
<td align="center">78.2</td>
<td align="center">58.7</td>
<td align="center">81.4</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-7B-Instruct</td>
<td align="center">8.3B</td>
<td align="center">70.9</td>
<td align="center">888</td>
<td align="center">68.1</td>
<td align="center">51.9</td>
<td align="center">58.0</td>
<td align="center">69.7</td>
<td align="center">82.2</td>
<td align="center">64.1</td>
<td align="center">84.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-8B</td>
<td align="center">8.1B</td>
<td align="center">68.1</td>
<td align="center">821</td>
<td align="center">64.5</td>
<td align="center">49.0</td>
<td align="center">56.2</td>
<td align="center">62.8</td>
<td align="center">82.5</td>
<td align="center">63.2</td>
<td align="center">84.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-2.6</td>
<td align="center">8.1B</td>
<td align="center">65.2</td>
<td align="center">852</td>
<td align="center">60.8</td>
<td align="center">48.1</td>
<td align="center">49.8</td>
<td align="center">60.0</td>
<td align="center">78.0</td>
<td align="center">57.5</td>
<td align="center">82.1</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o-2.6</td>
<td align="center">8.7B</td>
<td align="center">70.2</td>
<td align="center">889</td>
<td align="center">73.3</td>
<td align="center">51.1</td>
<td align="center">50.9</td>
<td align="center">67.2</td>
<td align="center">80.6</td>
<td align="center">63.3</td>
<td align="center">86.1</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-4.0</td>
<td align="center">4.1B</td>
<td align="center">69.0</td>
<td align="center">894</td>
<td align="center">66.9</td>
<td align="center">50.8</td>
<td align="center">51.2</td>
<td align="center">68.0</td>
<td align="center">79.7</td>
<td align="center">62.8</td>
<td align="center">82.9</td>
</tr>
</tbody>
</table>
</div>
</details>
<details>
<summary>Click to view single image results on ChartQA, MME, RealWorldQA, TextVQA, DocVQA, MathVision, DynaMath, WeMath, Object HalBench and MM Halbench. </summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th nowrap="nowrap" align="left">model</th>
<th nowrap="nowrap">Size</th>
<th nowrap="nowrap">ChartQA</th>
<th nowrap="nowrap">MME</th>
<th nowrap="nowrap">RealWorldQA</th>
<th nowrap="nowrap">TextVQA</th>
<th nowrap="nowrap">DocVQA</th>
<th nowrap="nowrap">MathVision</th>
<th nowrap="nowrap">DynaMath</th>
<th nowrap="nowrap">WeMath</th>
<th nowrap="nowrap" colspan="2">Obj Hal</th>
<th nowrap="nowrap" colspan="2">MM Hal</th>
</tr>
</thead>
<tbody>
<tr>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center">CHAIRs↓</td>
<td align="center">CHAIRi↓</td>
<td nowrap="nowrap" align="center">score avg@3↑</td>
<td nowrap="nowrap" align="center">hall rate avg@3↓</td>
</tr>
<tbody align="center">
<tr>
<td colspan="14" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4v-20240409</td>
<td align="center">-</td>
<td align="center">78.5</td>
<td align="center">1927</td>
<td align="center">61.4</td>
<td align="center">78.0</td>
<td align="center">88.4</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini-1.5-Pro</td>
<td align="center">-</td>
<td align="center">87.2</td>
<td align="center">-</td>
<td align="center">67.5</td>
<td align="center">78.8</td>
<td align="center">93.1</td>
<td align="center">41.0</td>
<td align="center">31.5</td>
<td align="center">50.5</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4.1-mini-20250414</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">45.3</td>
<td align="center">47.7</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Claude 3.5 Sonnet-20241022</td>
<td align="center">-</td>
<td align="center">90.8</td>
<td align="center">-</td>
<td align="center">60.1</td>
<td align="center">74.1</td>
<td align="center">95.2</td>
<td align="center">35.6</td>
<td align="center">35.7</td>
<td align="center">44.0</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td colspan="14" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-3B-Instruct</td>
<td align="center">3.8B</td>
<td align="center">84.0</td>
<td align="center">2157</td>
<td align="center">65.4</td>
<td align="center">79.3</td>
<td align="center">93.9</td>
<td align="center">21.9</td>
<td align="center">13.2</td>
<td align="center">22.9</td>
<td align="center">18.3</td>
<td align="center">10.8</td>
<td align="center">3.9 </td>
<td align="center">33.3 </td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-4B</td>
<td align="center">3.7B</td>
<td align="center">84.0</td>
<td align="center">2338</td>
<td align="center">64.3</td>
<td align="center">76.8</td>
<td align="center">91.6</td>
<td align="center">18.4</td>
<td align="center">15.2</td>
<td align="center">21.2</td>
<td align="center">13.7</td>
<td align="center">8.7</td>
<td align="center">3.2 </td>
<td align="center">46.5 </td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-7B-Instruct</td>
<td align="center">8.3B</td>
<td align="center">87.3</td>
<td align="center">2347</td>
<td align="center">68.5</td>
<td align="center">84.9</td>
<td align="center">95.7</td>
<td align="center">25.4</td>
<td align="center">21.8</td>
<td align="center">36.2</td>
<td align="center">13.3</td>
<td align="center">7.9</td>
<td align="center">4.1 </td>
<td align="center">31.6 </td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-8B</td>
<td align="center">8.1B</td>
<td align="center">84.8</td>
<td align="center">2344</td>
<td align="center">70.1</td>
<td align="center">79.1</td>
<td align="center">93.0</td>
<td align="center">17.0</td>
<td align="center">9.4</td>
<td align="center">23.5</td>
<td align="center">18.3</td>
<td align="center">11.6</td>
<td align="center">3.6 </td>
<td align="center">37.2</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-2.6</td>
<td align="center">8.1B</td>
<td align="center">79.4</td>
<td align="center">2348</td>
<td align="center">65.0</td>
<td align="center">80.1</td>
<td align="center">90.8</td>
<td align="center">17.5</td>
<td align="center">9.0</td>
<td align="center">20.4</td>
<td align="center">7.3</td>
<td align="center">4.7</td>
<td align="center">4.0 </td>
<td align="center">29.9 </td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o-2.6</td>
<td align="center">8.7B</td>
<td align="center">86.9</td>
<td align="center">2372</td>
<td align="center">68.1</td>
<td align="center">82.0</td>
<td align="center">93.5</td>
<td align="center">21.7</td>
<td align="center">10.4</td>
<td align="center">25.2</td>
<td align="center">6.3</td>
<td align="center">3.4</td>
<td align="center">4.1 </td>
<td align="center">31.3 </td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-4.0</td>
<td align="center">4.1B</td>
<td align="center">84.4</td>
<td align="center">2298</td>
<td align="center">68.5</td>
<td align="center">80.8</td>
<td align="center">92.9</td>
<td align="center">20.7</td>
<td align="center">14.2</td>
<td align="center">32.7</td>
<td align="center">6.3</td>
<td align="center">3.5</td>
<td align="center">4.1 </td>
<td align="center">29.2 </td>
</tr>
</tbody>
</table>
</div>
</details>
<details>
<summary>Click to view multi-image and video understanding results on Mantis, Blink and Video-MME. </summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th nowrap="nowrap" align="left">model</th>
<th nowrap="nowrap">Size</th>
<th nowrap="nowrap">Mantis</th>
<th nowrap="nowrap">Blink</th>
<th nowrap="nowrap" colspan="2" >Video-MME</th>
</tr>
</thead>
<tbody>
<tr>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center"></td>
<td align="center">wo subs</td>
<td align="center">w subs</td>
</tr>
<tbody align="center">
<tr>
<td colspan="6" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4v-20240409</td>
<td align="center">-</td>
<td align="center">62.7</td>
<td align="center">54.6</td>
<td align="center">59.9</td>
<td align="center">63.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini-1.5-Pro</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">59.1</td>
<td align="center">75.0</td>
<td align="center">81.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4o-20240513</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">68.0</td>
<td align="center">71.9</td>
<td align="center">77.2</td>
</tr>
<tr>
<td colspan="6" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-3B-Instruct</td>
<td align="center">3.8B</td>
<td align="center">-</td>
<td align="center">47.6</td>
<td align="center">61.5</td>
<td align="center">67.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-4B</td>
<td align="center">3.7B</td>
<td align="center">62.7</td>
<td align="center">50.8</td>
<td align="center">62.3</td>
<td align="center">63.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen2.5-VL-7B-Instruct</td>
<td align="center">8.3B</td>
<td align="center">-</td>
<td align="center">56.4</td>
<td align="center">65.1</td>
<td align="center">71.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2.5-8B</td>
<td align="center">8.1B</td>
<td align="center">67.7</td>
<td align="center">54.8</td>
<td align="center">64.2</td>
<td align="center">66.9</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-2.6</td>
<td align="center">8.1B</td>
<td align="center">69.1</td>
<td align="center">53.0</td>
<td align="center">60.9</td>
<td align="center">63.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-o-2.6</td>
<td align="center">8.7B</td>
<td align="center">71.9</td>
<td align="center">56.7</td>
<td align="center">63.9</td>
<td align="center">69.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V-4.0</td>
<td align="center">4.1B</td>
<td align="center">71.4</td>
<td align="center">54.0</td>
<td align="center">61.2</td>
<td align="center">65.8</td>
</tr>
</tbody>
</table>
</div>
</details>
### Examples <!-- omit in toc -->
<div style="display: flex; flex-direction: column; align-items: center;">
<img src="./assets/minicpmv4/minicpm-v-4-case.png" alt="math" style="margin-bottom: 5px;">
</div>
We deploy MiniCPM-V 4.0 on iPhone 16 Pro Max with [iOS demo](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/demo/ios_demo/ios.md). The demo video is the raw screen recording without edition.
<table align="center">
<p align="center">
<img src="./assets/minicpmv4/iphone_en.gif" width=45%/>
&nbsp;&nbsp;&nbsp;&nbsp;
<img src="./assets/minicpmv4/iphone_en_information_extraction.gif" width=45%/>
</p>
<p align="center">
<img src="./assets/minicpmv4/iphone_cn.gif" width=45%/>
&nbsp;&nbsp;&nbsp;&nbsp;
<img src="./assets/minicpmv4/iphone_cn_funny_points.gif" width=45%/>
</p>
</table>
</details>
## Legacy Models <!-- omit in toc -->
| Model | Introduction and Guidance |
|:----------------------|:-------------------:|
| MiniCPM-V 4.5 | [Document](./docs/minicpm_v4dot5_en.md) |
| MiniCPM-o 2.6 | [Document](./docs/minicpm_o2dot6_en.md) |
| MiniCPM-V 2.6 | [Document](./docs/minicpm_v2dot6_en.md) |
| MiniCPM-Llama3-V 2.5 | [Document](./docs/minicpm_llama3_v2dot5.md) |
| MiniCPM-V 2.0 | [Document](./docs/minicpm_v2.md) |
| MiniCPM-V 1.0 | [Document](./docs/minicpm_v1.md) |
| OmniLMM-12B | [Document](././docs/omnilmm_en.md) |
## MiniCPM-V & o Cookbook
Discover comprehensive, ready-to-deploy solutions for the MiniCPM-V and MiniCPM-o model series in our structured [Cookbook](https://github.com/OpenSQZ/MiniCPM-V-CookBook), which empowers developers to rapidly implement multimodal AI applications with integrated vision, speech, and live-streaming capabilities. Key features include:
**Easy Usage Documentation**
Our comprehensive [documentation website](https://minicpm-o.readthedocs.io/en/latest/index.html) presents every recipe in a clear, well-organized manner.
All features are displayed at a glance, making it easy for you to quickly find exactly what you need.
**Broad User Spectrum**
We support a wide range of users, from individuals to enterprises and researchers.
* **Individuals**: Enjoy effortless inference using Ollama ([V4](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/ollama/minicpm-v4_ollama.md), [o4.5](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/ollama/minicpm-o4_5_ollama.md)) and Llama.cpp ([V4](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/llama.cpp/minicpm-v4_llamacpp.md), [o4.5](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/llama.cpp/minicpm-o4_5_llamacpp.md)) with minimal setup.
* **Enterprises**: Achieve high-throughput, scalable performance with vLLM ([V4](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/vllm/minicpm-v4_vllm.md), [o4.5](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/vllm/minicpm-o4_5_vllm.md)) and SGLang ([V4](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/sglang/MiniCPM-v4_sglang.md), [o4.5](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/deployment/sglang/MiniCPM-o4_5_sglang.md)).
* **Researchers**: Leverage advanced frameworks including [Transformers](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/finetune/finetune_full.md), [LLaMA-Factory](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/finetune/finetune_llamafactory.md), [SWIFT](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/finetune/swift.md), and [Align-anything](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/finetune/align_anything.md) to enable flexible model development and cutting-edge experimentation.
**Versatile Deployment Scenarios**
Our ecosystem delivers optimal solution for a variety of hardware environments and deployment demands.
* **Web Demo**: Full-duplex real-time video interaction solution with high responsiveness and low latency. [WebRTC_Demo](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/demo/web_demo/WebRTC_Demo/README.md).
* **Quantized deployment**: Maximize efficiency and minimize resource consumption using [GGUF](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/quantization/gguf/minicpm-v4_gguf_quantize.md) and [BNB](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/quantization/bnb/minicpm-v4_bnb_quantize.md).
* **End devices**: Bring powerful AI experiences to [iPhone and iPad](https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/demo/ios_demo/ios.md), supporting offline and privacy-sensitive applications.
## Model Zoo
| Model | Device | Memory | &emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; Description | Download |
|:-----------|:--:|:-----------:|:-------------------|:---------------:|
| MiniCPM-o 4.5| GPU | 19 GB | The latest version, strong end-side multimodal performance for vision, speech and omni-modal live streaming on end-side devices. | [🤗](https://huggingface.co/openbmb/MiniCPM-o-4_5) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-o-4_5) |
| MiniCPM-o 4.5 gguf| GPU | 10 GB | The gguf version, lower memory usage and faster inference. | [🤗](https://huggingface.co/openbmb/MiniCPM-o-4_5-gguf) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-o-4_5-gguf) |
| MiniCPM-o 4.5 AWQ | GPU | 11 GB | The AWQ quantized version, lower GPU memory usage. | [🤗](https://huggingface.co/openbmb/MiniCPM-V-4_5-AWQ) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-o-4_5-awq) |
| MiniCPM-V 4.0| GPU | 9 GB | The latest version, strong end-side multimodal performance for single image, multi-image and video understanding. | [🤗](https://huggingface.co/openbmb/MiniCPM-V-4) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-4) |
| MiniCPM-V 4.0 gguf | CPU | 4 GB | The gguf version, lower memory usage and faster inference. | [🤗](https://huggingface.co/openbmb/MiniCPM-V-4-gguf) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-4-gguf) |
| MiniCPM-V 4.0 int4 | GPU | 5 GB | The int4 quantized version, lower GPU memory usage. | [🤗](https://huggingface.co/openbmb/MiniCPM-V-4-int4) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-4-int4) |
| MiniCPM-V 4.0 AWQ | GPU | 5 GB | The AWQ quantized version, lower GPU memory usage. | [🤗](https://huggingface.co/openbmb/MiniCPM-V-4-AWQ) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-4-AWQ) |
## Inference With Transformers
Inference using Hugging Face Transformers on NVIDIA GPUs. Please ensure `transformers==4.51.0` is installed, as other versions may have compatibility issues (under investigation). Requirements tested on Python 3.10:
- Without TTS or streaming inference:
```bash
pip install "transformers==4.51.0" accelerate "torch>=2.3.0,<=2.8.0" "torchaudio<=2.8.0" "minicpmo-utils>=1.0.2"
```
- With TTS or streaming inference:
```bash
pip install "transformers==4.51.0" accelerate "torch>=2.3.0,<=2.8.0" "torchaudio<=2.8.0" "minicpmo-utils[all]>=1.0.2"
```
### Model Initialization
<details>
<summary>Click to show model initialization code.</summary>
```python
import torch
from transformers import AutoModel
# Load omni model (default: init_vision=True, init_audio=True, init_tts=True)
# For vision-only model: set init_audio=False and init_tts=False
# For audio-only model: set init_vision=False
model = AutoModel.from_pretrained(
"openbmb/MiniCPM-o-4_5",
trust_remote_code=True,
attn_implementation="sdpa", # sdpa or flash_attention_2
torch_dtype=torch.bfloat16,
init_vision=True,
init_audio=True,
init_tts=True,
)
model.eval().cuda()
# Initialize TTS for audio output in chat or streaming mode
model.init_tts(streaming=False) # or streaming=True
# Convert simplex model to duplex mode
duplex_model = model.as_duplex()
# Convert duplex model back to simplex mode
simplex_model = duplex_model.as_simplex(reset_session=True)
```
</details>
### Duplex Omni Mode
Full-duplex streaming inference for real-time or recorded video conversations.
<details>
<summary>Click to show duplex omni mode code.</summary>
```python
import librosa
import torch
from minicpmo.utils import generate_duplex_video, get_video_frame_audio_segments
from transformers import AutoModel
# Load model and convert to duplex mode
model = AutoModel.from_pretrained(
"openbmb/MiniCPM-o-4_5",
trust_remote_code=True,
attn_implementation="sdpa", # or "flash_attention_2"
torch_dtype=torch.bfloat16,
)
model.eval().cuda()
model = model.as_duplex()
# Load video and reference audio
video_path = "assets/omni_duplex1.mp4"
ref_audio_path = "assets/HT_ref_audio.wav"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
# Extract video frames and audio segments
video_frames, audio_segments, stacked_frames = get_video_frame_audio_segments(
video_path, stack_frames=1, use_ffmpeg=True, adjust_audio_length=True
)
# Prepare duplex session with system prompt and voice reference
model.prepare(
prefix_system_prompt="Streaming Omni Conversation.",
ref_audio=ref_audio,
prompt_wav_path=ref_audio_path,
)
results_log = []
timed_output_audio = []
# Process each chunk in streaming fashion
for chunk_idx in range(len(audio_segments)):
audio_chunk = audio_segments[chunk_idx] if chunk_idx < len(audio_segments) else None
frame = video_frames[chunk_idx] if chunk_idx < len(video_frames) else None
frame_list = []
if frame is not None:
frame_list.append(frame)
if stacked_frames is not None and chunk_idx < len(stacked_frames) and stacked_frames[chunk_idx] is not None:
frame_list.append(stacked_frames[chunk_idx])
# Step 1: Streaming prefill
model.streaming_prefill(
audio_waveform=audio_chunk,
frame_list=frame_list,
max_slice_nums=1, # Increase for HD mode (e.g., [2, 1] for stacked frames)
batch_vision_feed=False, # Set True for faster processing
)
# Step 2: Streaming generate
result = model.streaming_generate(
prompt_wav_path=ref_audio_path,
max_new_speak_tokens_per_chunk=20,
decode_mode="sampling",
)
if result["audio_waveform"] is not None:
timed_output_audio.append((chunk_idx, result["audio_waveform"]))
chunk_result = {
"chunk_idx": chunk_idx,
"is_listen": result["is_listen"],
"text": result["text"],
"end_of_turn": result["end_of_turn"],
"current_time": result["current_time"],
"audio_length": len(result["audio_waveform"]) if result["audio_waveform"] is not None else 0,
}
results_log.append(chunk_result)
print("listen..." if result["is_listen"] else f"speak> {result['text']}")
# Generate output video with AI responses
generate_duplex_video(
video_path=video_path,
output_video_path="duplex_output.mp4",
results_log=results_log,
timed_output_audio=timed_output_audio,
output_sample_rate=24000,
)
```
</details>
### Simplex Omni Mode
We provide two inference modes: chat and streaming.
#### Chat Inference <!-- omit in toc -->
<details>
<summary>Click to show chat inference code.</summary>
```python
from minicpmo.utils import get_video_frame_audio_segments
model = ...
model.init_tts(streaming=False)
video_path = "assets/Skiing.mp4"
# Optional: Set reference audio for voice cloning
ref_audio_path = "assets/HT_ref_audio.wav"
sys_msg = model.get_sys_prompt(ref_audio=ref_audio_path, mode="omni", language="en")
# Use stack_frames=5 for high refresh rate mode
video_frames, audio_segments, stacked_frames = get_video_frame_audio_segments(video_path, stack_frames=1)
omni_contents = []
for i in range(len(video_frames)):
omni_contents.append(video_frames[i])
omni_contents.append(audio_segments[i])
if stacked_frames is not None and stacked_frames[i] is not None:
omni_contents.append(stacked_frames[i])
msg = {"role": "user", "content": omni_contents}
msgs = [sys_msg, msg]
# Set generate_audio=True and output_audio_path to save TTS output
generate_audio = True
output_audio_path = "output.wav"
res = model.chat(
msgs=msgs,
max_new_tokens=4096,
do_sample=True,
temperature=0.7,
use_tts_template=True,
enable_thinking=False,
omni_mode=True, # Required for omni inference
generate_audio=generate_audio,
output_audio_path=output_audio_path,
max_slice_nums=1, # Increase for HD mode
)
print(res)
# Example output: "The person in the picture is skiing down a snowy mountain slope."
# import IPython
# IPython.display.Audio("output.wav")
```
</details>
#### Streaming Inference <!-- omit in toc -->
<details>
<summary>Click to show streaming inference code.</summary>
```python
import librosa
import numpy as np
import soundfile as sf
import torch
from minicpmo.utils import get_video_frame_audio_segments
model = ...
model.init_tts(streaming=True)
# Reset session for a new conversation (clears KV cache)
model.reset_session()
# Optional: Load reference audio for voice cloning
ref_audio_path = "assets/HT_ref_audio.wav"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
model.init_token2wav_cache(ref_audio)
session_id = "demo"
# Extract video frames and audio segments (use stack_frames=5 for high refresh rate mode)
video_path = "assets/Skiing.mp4"
video_frames, audio_segments, stacked_frames = get_video_frame_audio_segments(video_path, stack_frames=1)
# Build omni contents list
omni_contents = []
for i in range(len(video_frames)):
omni_contents.append(video_frames[i])
omni_contents.append(audio_segments[i])
if stacked_frames is not None and stacked_frames[i] is not None:
omni_contents.append(stacked_frames[i])
generate_audio = False
output_audio_path = "output.wav"
# Step 1: Prefill system prompt
sys_msg = model.get_sys_prompt(ref_audio=ref_audio, mode="omni", language="en")
model.streaming_prefill(session_id=session_id, msgs=[sys_msg])
# Step 2: Prefill omni chunks (is_last_chunk=True only for the last audio chunk)
audio_indices = [i for i, c in enumerate(omni_contents) if isinstance(c, np.ndarray)]
last_audio_idx = audio_indices[-1] if audio_indices else -1
for idx, content in enumerate(omni_contents):
is_last_audio_chunk = idx == last_audio_idx
msgs = [{"role": "user", "content": [content]}]
model.streaming_prefill(session_id=session_id, msgs=msgs, omni_mode=True, is_last_chunk=is_last_audio_chunk)
# Step 3: Generate response
iter_gen = model.streaming_generate(
session_id=session_id,
generate_audio=generate_audio,
use_tts_template=True,
enable_thinking=False,
do_sample=True,
)
audios = []
text = ""
if generate_audio:
for wav_chunk, text_chunk in iter_gen:
audios.append(wav_chunk)
text += text_chunk
generated_waveform = torch.cat(audios, dim=-1)[0]
sf.write(output_audio_path, generated_waveform.cpu().numpy(), samplerate=24000)
print("Text:", text)
print("Audio saved to output.wav")
else:
for text_chunk, is_finished in iter_gen:
text += text_chunk
print("Text:", text)
```
</details>
### Simplex Realtime Speech Conversation Mode <!-- omit in toc -->
<details>
<summary>Click to show simplex mode realtime speech conversation API usage.</summary>
First, make sure you have all dependencies, especially `minicpmo-utils[all]>=1.0.2`:
```bash
pip install "transformers==4.51.0" accelerate "torch>=2.3.0,<=2.8.0" "torchaudio<=2.8.0" "minicpmo-utils[all]>=1.0.2"
```
```python
import librosa
import numpy as np
import torch
import soundfile as sf
model = ...
# Set reference audio for voice style
ref_audio_path = "ref_audio_path"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
# Example system msg for English Conversation
sys_msg = {
"role": "system",
"content": [
"Clone the voice in the provided audio prompt.",
ref_audio,
"Please assist users while maintaining this voice style. Please answer the user's questions seriously and in a high quality. Please chat with the user in a highly human-like and oral style. You are a helpful assistant developed by ModelBest: MiniCPM-Omni"
]
}
# Example system msg for Chinese Conversation
sys_msg = {
"role": "system",
"content": [
"模仿输入音频中的声音特征。",
ref_audio,
"你的任务是用这种声音模式来当一个助手。请认真、高质量地回复用户的问题。请用高自然度的方式和用户聊天。你是由面壁智能开发的人工智能助手：面壁小钢炮。"
]
}
# You can use each type of system prompt mentioned above in streaming speech conversation
# Reset state
model.init_tts(streaming=True)
model.reset_session(reset_token2wav_cache=True)
model.init_token2wav_cache(prompt_speech_16k=ref_audio)
session_id = "demo"
# First, prefill system turn
model.streaming_prefill(
session_id=session_id,
msgs=[sys_msg],
omni_mode=False,
is_last_chunk=True,
)
# Here we simulate realtime speech conversation by splitting whole user input audio into chunks of 1s.
user_audio, _ = librosa.load("user_audio.wav", sr=16000, mono=True)
IN_SAMPLE_RATE = 16000 # input audio sample rate, fixed value
CHUNK_SAMPLES = IN_SAMPLE_RATE # sample
OUT_SAMPLE_RATE = 24000 # output audio sample rate, fixed value
total_samples = len(user_audio)
num_chunks = (total_samples + CHUNK_SAMPLES - 1) // CHUNK_SAMPLES
for chunk_idx in range(num_chunks):
start = chunk_idx * CHUNK_SAMPLES
end = min((chunk_idx + 1) * CHUNK_SAMPLES, total_samples)
chunk_audio = user_audio[start:end]
is_last_chunk = (chunk_idx == num_chunks - 1)
user_msg = {"role": "user", "content": [chunk_audio]}
# For each 1s audio chunk, perform streaming_prefill once to reduce first-token latency
model.streaming_prefill(
session_id=session_id,
msgs=[user_msg],
omni_mode=False,
is_last_chunk=is_last_chunk,
)
# Let model generate response in a streaming manner
generate_audio = True
iter_gen = model.streaming_generate(
session_id=session_id,
generate_audio=generate_audio,
use_tts_template=True,
enable_thinking=False,
do_sample=True,
max_new_tokens=512,
length_penalty=1.1, # For realtime speech conversation mode, we suggest length_penalty=1.1 to improve response content
)
audios = []
text = ""
if generate_audio:
for wav_chunk, text_chunk in iter_gen:
audios.append(wav_chunk)
text += text_chunk
generated_waveform = torch.cat(audios, dim=-1)[0]
sf.write(output_audio_path, generated_waveform.cpu().numpy(), samplerate=24000)
print("Text:", text)
print("Audio saved to output.wav")
else:
for text_chunk, is_finished in iter_gen:
text += text_chunk
print("Text:", text)
# Now we can prefill the following user turns and generate next turn response...
```
</details>
#### Speech Conversation as a Versatile and Vibe AI Assistant <!-- omit in toc -->
Built on carefully designed post-training data and professional voice-actor recordings, `MiniCPM-o-4.5` can also function as an AI voice assistant. It delivers high-quality spoken interaction out of the box. It produces a sweet and expressive voice with natural prosody, including appropriate rhythm, stress, and pauses, giving a strong sense of liveliness in casual conversation. It also supports storytelling and narrative speech with coherent and engaging delivery. Moreover, it enables advanced voice instruction control. like emotional tone, word-level emphasis.
<details>
<summary>Click to show AI assistant conversation code.</summary>
```python
import librosa
# Set reference audio for voice style
ref_audio_path = "assets/HT_ref_audio.wav"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
# For Chinese Conversation
sys_msg = {
"role": "system",
"content": [
"模仿输入音频中的声音特征。",
ref_audio,
"你的任务是用这种声音模式来当一个助手。请认真、高质量地回复用户的问题。请用高自然度的方式和用户聊天。你是由面壁智能开发的人工智能助手：面壁小钢炮。"
]
}
# For English Conversation
sys_msg = {
"role": "system",
"content": [
"Clone the voice in the provided audio prompt.",
ref_audio,
"Please assist users while maintaining this voice style. Please answer the user's questions seriously and in a high quality. Please chat with the user in a highly human-like and oral style. You are a helpful assistant developed by ModelBest: MiniCPM-Omni."
]
}
```
</details>
#### General Speech Conversation with Custom Voice and Custom System Profile <!-- omit in toc -->
MiniCPM-o-4.5 can role-play as a specific character based on an audio prompt and text profile prompt. It mimics the character's voice and adopts their language style in text responses. It also follows profile defined in text profile. In this mode, MiniCPM-o-4.5 sounds **more natural and human-like**.
<details>
<summary>Click to show custom voice conversation code.</summary>
```python
import librosa
# Set reference audio for voice cloning
ref_audio_path = "assets/system_ref_audio.wav"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
# For English conversation with text profile
sys_msg = {
"role": "system",
"content": [
"Clone the voice in the provided audio prompt.",
ref_audio,
"Please chat with the user in a highly human-like and oral style." + "You are Elon Musk, CEO of Tesla and SpaceX. You speak directly and casually, often with dry humor. You're passionate about Mars, sustainable energy, and pushing humanity forward. Speak bluntly with occasional dark humor. Use simple logic and don't sugarcoat things. Don't be diplomatic. Say what you actually think, even if it's controversial. Keep responses around 100 words. Don't ramble."
]
}
# For English conversation with no text profile
sys_msg = {
"role": "system",
"content": [
"Clone the voice in the provided audio prompt.",
ref_audio,
"Your task is to be a helpful assistant using this voice pattern. Please answer the user's questions seriously and in a high quality. Please chat with the user in a high naturalness style."
]
}
# For Chinese Conversation with no text profile
sys_msg = {
"role": "system",
"content": [
"根据输入的音频提示生成相似的语音。",
librosa.load("assets/system_ref_audio_2.wav", sr=16000, mono=True)[0],
"作为助手，你将使用这种声音风格说话。 请认真、高质量地回复用户的问题。 请用高自然度的方式和用户聊天。"
]
}
# For Chinese Conversation with text profile
sys_msg = {
"role": "system",
"content": [
"根据输入的音频提示生成相似的语音。",
ref_audio,
"你是一个具有以上声音风格的AI助手。请用高拟人度、口语化的方式和用户聊天。" + "你是一名心理咨询师兼播客主理人，热爱创作与深度对话。你性格细腻、富有共情力，善于从个人经历中提炼哲思。语言风格兼具理性与诗意，常以隐喻表达内在体验。"
]
}
```
</details>
### Speech and Audio Mode <!-- omit in toc -->
#### Zero-shot Text-to-speech (TTS) <!-- omit in toc -->
`MiniCPM-o-4.5` supports zero-shot text-to-speech (TTS). In this mode, the model functions as a highly-natural TTS system that can replicate a reference voice.
<details>
<summary>Click to show TTS code.</summary>
```python
import librosa
model = ...
model.init_tts(streaming=False)
# For both Chinese and English
ref_audio_path = "assets/HT_ref_audio.wav"
ref_audio, _ = librosa.load(ref_audio_path, sr=16000, mono=True)
sys_msg = {"role": "system", "content": [
"模仿音频样本的音色并生成新的内容。",
ref_audio,
"请用这种声音风格来为用户提供帮助。 直接作答，不要有冗余内容"
]}
# For English
user_msg = {
"role": "user",
"content": [
"请朗读以下内容。" + " " + "I have a wrap up that I want to offer you now, a conclusion to our work together."
]
}
# For Chinese
user_msg = {
"role": "user",
"content": [
"请朗读以下内容。" + " " + "你好，欢迎来到艾米说科幻，我是艾米。"
]
}
msgs = [sys_msg, user_msg]
res = model.chat(
msgs=msgs,
do_sample=True,
max_new_tokens=512,
use_tts_template=True,
generate_audio=True,
temperature=0.1,
output_audio_path="result_voice_cloning.wav",
)
```
</details>
#### Mimick <!-- omit in toc -->
The `Mimick` task evaluates a model's end-to-end speech modeling capability. The model takes audio input, transcribes it, and reconstructs the original audio with high fidelity, preserving detailed acoustic, paralinguistic, and semantic information. Higher similarity between the reconstructed and original audio indicates stronger end-to-end speech modeling capability.
<details>
<summary>Click to show mimick code.</summary>
```python
import librosa
model = ...
model.init_tts(streaming=False)
system_prompt = "You are a helpful assistant. You can accept video, audio, and text input and output voice and text. Respond with just the answer, no redundancy."
mimick_prompt = "Please repeat the following speech in the appropriate language."
audio_input, _ = librosa.load("assets/Trump_WEF_2018_10s.mp3", sr=16000, mono=True)
msgs = [
{"role": "system", "content": [system_prompt]},
{"role": "user", "content": [mimick_prompt, audio_input]}
]
res = model.chat(
msgs=msgs,
do_sample=True,
max_new_tokens=512,
use_tts_template=True,
temperature=0.1,
generate_audio=True,
output_audio_path="output_mimick.wav",
)
```
</details>
#### Addressing Various Audio Understanding Tasks <!-- omit in toc -->
`MiniCPM-o-4.5` can also handle various audio understanding tasks, such as ASR, speaker analysis, general audio captioning, and sound scene tagging.
For audio-to-text tasks, you can use the following prompts:
- ASR (Chinese, or AST EN→ZH): `请仔细听这段音频片段，并将其内容逐字记录。`
- ASR (English, or AST ZH→EN): `Please listen to the audio snippet carefully and transcribe the content.`
- Speaker Analysis: `Based on the speaker's content, speculate on their gender, condition, age range, and health status.`
- General Audio Caption: `Summarize the main content of the audio.`
- Sound Scene Tagging: `Utilize one keyword to convey the audio's content or the associated scene.`
<details>
<summary>Click to show audio understanding code.</summary>
```python
import librosa
model = ...
model.init_tts(streaming=False)
# Load the audio to be transcribed/analyzed
audio_input, _ = librosa.load("assets/Trump_WEF_2018_10s.mp3", sr=16000, mono=True)
# Choose a task prompt (see above for options)
task_prompt = "Please listen to the audio snippet carefully and transcribe the content.\n"
msgs = [{"role": "user", "content": [task_prompt, audio_input]}]
res = model.chat(
msgs=msgs,
do_sample=True,
max_new_tokens=512,
use_tts_template=True,
generate_audio=True,
temperature=0.3,
output_audio_path="result_audio_understanding.wav",
)
print(res)
```
</details>
### Visual Understanding
`MiniCPM-o-4.5` shares the same inference methods as `MiniCPM-V-4.5`.
#### Chat with Single Image <!-- omit in toc -->
<details>
<summary>Click to show single image chat code.</summary>
```python
import torch
from PIL import Image
from transformers import AutoModel
model = AutoModel.from_pretrained(
"openbmb/MiniCPM-o-4_5",
trust_remote_code=True,
attn_implementation="sdpa", # or "flash_attention_2"
torch_dtype=torch.bfloat16,
init_vision=True,
init_audio=False,
init_tts=False,
)
model.eval().cuda()
image = Image.open("assets/fossil.png").convert("RGB")
question = "What is in the image?"
msgs = [{"role": "user", "content": [image, question]}]
res = model.chat(msgs=msgs, use_tts_template=False)
print(res)
```
</details>
#### Chat with Multiple Images <!-- omit in toc -->
<details>
<summary>Click to show Python code for multi-image input.</summary>
```python
import torch
from PIL import Image
from transformers import AutoModel
model = ...
image1 = Image.open("assets/highway.png").convert("RGB")
image2 = Image.open("assets/fossil.png").convert("RGB")
question = "Compare image 1 and image 2, tell me about the differences between them."
msgs = [{"role": "user", "content": [image1, image2, question]}]
answer = model.chat(msgs=msgs, use_tts_template=False, enable_thinking=False)
print(answer)
```
</details>
#### In-Context Few-Shot Learning <!-- omit in toc -->
<details>
<summary>Click to show Python code for few-shot learning.</summary>
```python
from PIL import Image
model = ...
question = "production date"
image1 = Image.open("example1.jpg").convert("RGB")
answer1 = "2023.08.04"
image2 = Image.open("example2.jpg").convert("RGB")
answer2 = "2007.04.24"
image_test = Image.open("test.jpg").convert("RGB")
msgs = [
{"role": "user", "content": [image1, question]},
{"role": "assistant", "content": [answer1]},
{"role": "user", "content": [image2, question]},
{"role": "assistant", "content": [answer2]},
{"role": "user", "content": [image_test, question]},
]
answer = model.chat(msgs=msgs, use_tts_template=False, enable_thinking=False)
print(answer)
```
</details>
#### Chat with Video <!-- omit in toc -->
<details>
<summary>Click to show Python code for video input.</summary>
```python
import torch
from minicpmo.utils import get_video_frame_audio_segments
from transformers import AutoModel
model = ...
video_path = "assets/Skiing.mp4"
video_frames, _, _ = get_video_frame_audio_segments(video_path)
print("num frames:", len(video_frames))
question = "Describe the video"
msgs = [{"role": "user", "content": video_frames + [question]}]
answer = model.chat(
msgs=msgs,
max_new_tokens=128,
use_image_id=False,
max_slice_nums=1,
use_tts_template=False,
enable_thinking=False, # Set True to enable thinking mode
)
print(answer)
```
</details>
### Structured Content Input
<details>
<summary>Click to show structured content input details.</summary>
The `chat` method accepts message content in two formats:
**Native format** – pass Python objects directly:
```python
msgs = [{"role": "user", "content": [pil_image, audio_ndarray, "Describe this."]}]
```
**OpenAI-compatible format** – use structured dictionaries:
```python
msgs = [
{
"role": "user",
"content": [
{"type": "image_url", "image_url": {"url": "/path/to/image.jpg"}},
{"type": "audio_url", "audio_url": {"url": "/path/to/audio.wav"}},
{"type": "video_url", "video_url": {"url": "/path/to/video.mp4", "use_audio": True}},
{"type": "text", "text": "Describe this."}
]
}
]
```
**Supported types:**
| Type | Input | Converts to |
|------|-------|-------------|
| `text` | `{"type": "text", "text": "..."}` | `str` |
| `image_url` | `{"type": "image_url", "image_url": {"url": "..."}}` | `PIL.Image` |
| `audio_url` | `{"type": "audio_url", "audio_url": {"url": "..."}}` | `np.ndarray` (16kHz mono) |
| `video_url` | `{"type": "video_url", "video_url": {"url": "...", "stack_frames": 1, "use_audio": True}}` | `List[Image, ndarray, ...]` |
- **URL sources**: local file paths or `http://`/`https://` URLs
- **Mixed formats**: native objects and structured dicts can be combined in the same content list
</details>
## Supported Frameworks
### FlagOS
To enable large-scale deployment across different AI chips, Beijing Zhiyuan Research Institute, together with numerous research institutions, chip manufacturers, system vendors, and algorithm and software organizations both domestically and internationally, jointly initiated and established the FlagOS Open Source Community.
The FlagOS community is dedicated to building a unified, open-source system software stack for various AI chips, encompassing core open-source projects such as a large-scale operator library, a unified AI compiler, parallel training and inference frameworks, and a unified communication library. It aims to create an open technology ecosystem connecting the “model-system-chip” layers. By enabling “develop once, deploy across chips”, FlagOS unlocks the computational potential of hardware, breaks down the ecosystem silos between different chip software stacks, and effectively reduces migration costs for developers. The FlagOS community fosters an AI hardware and software ecosystem, overcomes single-vendor closed-source monopolies, promotes widespread deployment of AI hardware technologies, and is committed to rooted in China while embracing global collaboration.
Official website: https://flagos.io.
<details>
<summary>Click to show FlagOS details.</summary>
#### FlagOS: Supporting Multiple AI Chips <!-- omit in toc -->
Thanks to FlagOS’s unified multi-chip AI system software stack, MiniCPM-o 4.5 was adapted to 6 different AI chips in an extremely short time. Currently, the multi-chip version of MiniCPM-o 4.5 has been released on FlagRelease, FlagOS’s platform for automatic migration, adaptation, and deployment of large models across multi-architecture AI chips. Details are as follows:
| Vendor | ModelScope | Huggingface |
|:----------------|:------------:|:------------:|
| Nvidia | [MiniCPM-o-4.5-nvidia-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS) | [MiniCPM-o-4.5-nvidia-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS) |
| Hygon-BW1000 | [MiniCPM-o-4.5-hygon-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-hygon-FlagOS) | [MiniCPM-o-4.5-hygon-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-hygon-FlagOS) |
| Metax-C550 | [MiniCPM-o-4.5-metax-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-metax-FlagOS) | [MiniCPM-o-4.5-metax-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-metax-FlagOS) |
| Iluvatar-BIV150 | [MiniCPM-o-4.5-iluvatar-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-iluvatar-FlagOS) | [MiniCPM-o-4.5-iluvatar-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-iluvatar-FlagOS) |
| Ascend-A3 | [MiniCPM-o-4.5-ascend-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-ascend-FlagOS) | [MiniCPM-o-4.5-ascend-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-ascend-FlagOS) |
| Zhenwu-810E | [MiniCPM-o-4.5-zhenwu-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-zhenwu-FlagOS) | [MiniCPM-o-4.5-zhenwu-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-zhenwu-FlagOS) |
##### Comprehensive Evaluation <!-- omit in toc -->
###### Transformers-FlagOS version <!-- omit in toc -->
Accuracy Difference between `USE_FLAGOS=1` on multi-backend and `USE_FLAGOS=0` on Nvidia-CUDA
| Metrics | FlagOS Backend | Difference with Nvidia-CUDA |
|:-------------------------|:---------------:|:---------------------------:|
| Video-MME 0-shot avg@1 ↑ | Nvidia | 0.33% |
| Video-MME 0-shot avg@1 ↑ | Hygon-BW1000 | 0.17% |
| Video-MME 0-shot avg@1 ↑ | Ascend-A3 | 0.50% |
| Video-MME 0-shot avg@1 ↑ | Iluvatar-BIV150 | 1.83% |
| Video-MME 0-shot avg@1 ↑ | Metax-C550 | 0.75% |
###### VLLM-FlagOS version <!-- omit in toc -->
Accuracy Difference between `USE_FLAGGEMS=1 FLAGCX_PATH=/workspace/FlagCX` on Nvidia or `USE_FLAGGEMS=1` on ZHENW 810E, and launching vllm server directly on Nvidia
| Metrics (avg@1) | Difference between Nvidia-FlagOS and Nvidia-CUDA | Difference between Zhenwu-FlagOS and Nvidia-CUDA |
|:--------------------|:------------------------------------------------:|:------------------------------------------------:|
| CMMMU ↑ | 0.72% | 3.5% |
| MMMU ↑ | 1.44% | 1.18% |
| MMMU_Pro_standard ↑ | 0.83% | 0.22% |
| MM-Vet v2 ↑ | 0.46% | 1.33% |
| OCRBench ↑ | 0.10% | 1% |
| CII-Bench ↑ | 0.40% | 0.13% |
| Blink ↑ | 1.90% | 2.19% |
#### FlagOS Usage <!-- omit in toc -->
##### FlagOS Performance Acceleration on Nvidia <!-- omit in toc -->
On the Transformers version, under the premise of precision alignment between the CUDA and FlagOS ecosystems, FlagOS achieves a 6% performance improvement in total task execution time compared to CUDA.
###### From FlagRelease【Recommendation】 <!-- omit in toc -->
FlagRelease is a platform developed by the FlagOS team for automatic migration, adaptation, and deployment of large models across multi-architecture AI chips. The multi-chip version of MiniCPM-o 4.5 has already been released on FlagRelease. All necessary software packages are pre-installed on the platform, so users do not need to install anything.
- FlagRelease Image Key Versions
| Component | Version |
|:------------------------|:------------------------------------|
| Accelerator Card Driver | 570.158.01 |
| CUDA SDK Build | cuda_13.0.r13.0/compiler.36424714_0 |
| FlagTree | 0.4.0+3.5 |
| FlagGems | 4.2.1rc0 |
| vllm & vllm-plugin-fl | 0.13.0 + vllm_fl 0.0.0 |
| FlagCX | 0.1.0 |
- FlagRelease Quick Start
| Vendor | ModelScope | Huggingface |
|:-----------|:------------:|:------------:|
| Nvidia | [MiniCPM-o-4.5-nvidia-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS) | [MiniCPM-o-4.5-nvidia-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS) |
| Hygon-BW1000 | [MiniCPM-o-4.5-hygon-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-hygon-FlagOS) | [MiniCPM-o-4.5-hygon-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-hygon-FlagOS) |
| Metax-C550 | [MiniCPM-o-4.5-metax-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-metax-FlagOS) | [MiniCPM-o-4.5-metax-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-metax-FlagOS) |
| Iluvatar-BIV150 | [MiniCPM-o-4.5-iluvatar-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-iluvatar-FlagOS) | [MiniCPM-o-4.5-iluvatar-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-iluvatar-FlagOS) |
| Ascend-A3 | [MiniCPM-o-4.5-ascend-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-ascend-FlagOS) | [MiniCPM-o-4.5-ascend-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-ascend-FlagOS) |
| Zhenwu-810E | [MiniCPM-o-4.5-zhenwu-FlagOS](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-zhenwu-FlagOS) | [MiniCPM-o-4.5-zhenwu-FlagOS](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-zhenwu-FlagOS) |
##### From Scratch <!-- omit in toc -->
- Dependencies: Python 3.12, GLIBC 2.39, GLIBCXX 3.4.33, CXXABI 1.3.15
###### Transformers <!-- omit in toc -->
- Installing the FlagOS Operator Library
Official Repository: https://github.com/flagos-ai/FlagGems
```shell
pip install flag-gems==4.2.1rc0
```
- Installing the FlagOS Compiler
Official Repository: https://github.com/flagos-ai/flagtree
Quick Reference for Core Dependency Versions: https://github.com/flagos-ai/FlagTree/blob/main/documents/build.md#tips-for-building
```shell
pip uninstall triton
python3 -m pip install flagtree==0.4.0+3.5 --index-url=https://resource.flagos.net/repository/flagos-pypi-hosted/simple --trusted-host=https://resource.flagos.net
```
- Activating Acceleration
Add `USE_FLAGOS=1` before the command for the task you want to run. For example, when you run:
```shell
python3 generate_speech_from_video.py
```
To use the MiniCPM-o-4.5 model to generate spoken responses from video content, you can:
```shell
USE_FLAGOS=1 python3 generate_speech_from_video.py
```
to accelerate this process with FlagOS.
###### Vllm Version <!-- omit in toc -->
- Installing the FlagOS Operator Library
Official Repository: https://github.com/flagos-ai/FlagGems
```shell
pip install flag-gems==4.2.1rc0
pip install triton==3.5.1
```
- Activating Acceleration
Add `USE_FLAGOS=1` before the command for the task you want to run. For example, when you run:
```shell
vllm serve ${model_path} --dtype auto --gpu_memory_utilization 0.9 --trust-remote-code --max-num-batched-tokens 2048 --served-model-name cpmo --port ${Port}
```
To start the MiniCPM-o-4.5 server, you can:
```shell
USE_FLAGOS=1 vllm serve ${model_path} --dtype auto --gpu_memory_utilization 0.9 --trust-remote-code --max-num-batched-tokens 2048 --served-model-name cpmo --port ${Port}
```
to accelerate this process with FlagOS.
#### Using FlagOS Unified Multi-Chip Backend Plugin <!-- omit in toc -->
[vllm-plugin-FL](https://github.com/flagos-ai/vllm-plugin-FL) is a plugin built for the vLLM inference/service framework. Developed on top of FlagOS’s unified multi-chip backend, it is designed to extend vLLM’s capabilities and performance across a variety of hardware environments.
##### Using vllm-plugin-FL <!-- omit in toc -->
| Vendor | From Scratch | From FlagRelease |
|:-------|:-------------|:----------------|
| Nvidia | [vllm-plugin-FL/MiniCPM-o-4.5](https://github.com/flagos-ai/vllm-plugin-FL/blob/main/examples/minicpm/README.md) | [MiniCPM-o-4.5-ModelScope](https://modelscope.cn/models/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS), [MiniCPM-o-4.5-HuggingFace](https://huggingface.co/FlagRelease/MiniCPM-o-4.5-nvidia-FlagOS) |
</details>
### vLLM, SGLang, llama.cpp, Ollama
We support inference with vLLM, SGLang, llama.cpp and Ollama. Refer to our [Cookbook](https://github.com/OpenSQZ/MiniCPM-V-Cookbook) for more details.
### LLaMA-Factory, SWIFT
We support fine-tuning with LLaMA-Factory, SWIFT. Refer to our [Cookbook](https://github.com/OpenSQZ/MiniCPM-V-Cookbook) for more details.
## Awesome work using MiniCPM-V & MiniCPM-o
- [text-extract-api](https://github.com/CatchTheTornado/text-extract-api): Document extraction API using OCRs and Ollama supported models ![GitHub Repo stars](https://img.shields.io/github/stars/CatchTheTornado/text-extract-api)
- [comfyui_LLM_party](https://github.com/heshengtao/comfyui_LLM_party): Build LLM workflows and integrate into existing image workflows ![GitHub Repo stars](https://img.shields.io/github/stars/heshengtao/comfyui_LLM_party)
- [Ollama-OCR](https://github.com/imanoop7/Ollama-OCR): OCR package uses vlms through Ollama to extract text from images and PDF ![GitHub Repo stars](https://img.shields.io/github/stars/imanoop7/Ollama-OCR)
- [comfyui-mixlab-nodes](https://github.com/MixLabPro/comfyui-mixlab-nodes): ComfyUI node suite supports Workflow-to-APP、GPT&3D and more ![GitHub Repo stars](https://img.shields.io/github/stars/MixLabPro/comfyui-mixlab-nodes)
- [OpenAvatarChat](https://github.com/HumanAIGC-Engineering/OpenAvatarChat): Interactive digital human conversation implementation on single PC ![GitHub Repo stars](https://img.shields.io/github/stars/HumanAIGC-Engineering/OpenAvatarChat)
- [pensieve](https://github.com/arkohut/pensieve): A privacy-focused passive recording project by recording screen content ![GitHub Repo stars](https://img.shields.io/github/stars/arkohut/pensieve)
- [paperless-gpt](https://github.com/icereed/paperless-gpt): Use LLMs to handle paperless-ngx, AI-powered titles, tags and OCR ![GitHub Repo stars](https://img.shields.io/github/stars/icereed/paperless-gpt)
- [Neuro](https://github.com/kimjammer/Neuro): A recreation of Neuro-Sama, but running on local models on consumer hardware ![GitHub Repo stars](https://img.shields.io/github/stars/kimjammer/Neuro)
## Limitations
As an experimental trial, we find MiniCPM-o 4.5 has notable limitations worth further investigation and improvement.
- **Foundation Capability.** The full-duplex omni-modality live streaminig capability still needs improvement in its foundation capability.
- **Unstable Speech Output in Omni Mode.** Speech synthesis can mispronounce characters in full-duplex omni-modal live streaminig mode.
- **Mixed Language.** The model can sometimes respond with mixed English and Chinese in speech and omni mode.
- **High-latency on Web Demo.** Users may experience unusual high-latency or even miss part of model output fragments when using our web demo hosted on overseas servers. We recommend deploying the demo locally or with good network connections.
## Model License <!-- omit in toc -->
* The MiniCPM-o/V model weights and code are open-sourced under the [Apache-2.0](https://github.com/OpenBMB/MiniCPM-V/blob/main/LICENSE) license.
* To help us better understand and support our users, we would deeply appreciate it if you could consider optionally filling out a brief registration ["questionnaire"](https://modelbest.feishu.cn/share/base/form/shrcnpV5ZT9EJ6xYjh3Kx0J6v8g).
## Statement <!-- omit in toc -->
As MLLMs, MiniCPM-o/V models generate content by learning a large number of multimodal corpora, but they cannot comprehend, express personal opinions, or make value judgements. Anything generated by MiniCPM-o/V models does not represent the views and positions of the model developers
We will not be liable for any problems arising from the use of MiniCPM-o/V models, including but not limited to data security issues, risk of public opinion, or any risks and problems arising from the misdirection, misuse, dissemination, or misuse of the model.
## Acknowledgements
We would like to thank the following projects:
* [Qwen3](https://huggingface.co/Qwen/Qwen3-8B) for providing language backbone
* [SigLIP2](https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/image_text/README_siglip2.md) for providing vision understanding module
* [Whisper](https://github.com/openai/whisper) for providing audio and speech understanding module
* [CosyVoice2](https://github.com/FunAudioLLM/CosyVoice) and [Step-Audio2](https://github.com/stepfun-ai/Step-Audio2) for providing speech tokenizer and high-efficiency Token2Wav module.
* [Transformers](https://github.com/huggingface/transformers)
## Institutions <!-- omit in toc -->
This project is developed by the following institutions:
- <img src="assets/thunlp.png" width="28px"> [THUNLP](https://nlp.csai.tsinghua.edu.cn/)
- <img src="assets/modelbest.png" width="28px"> [ModelBest](https://modelbest.cn/)
## 🌟 Star History <!-- omit in toc -->
<table align="center">
<p align="center">
<img src="assets/star-history-25-09-02.png"/>
</p>
</table>
<!-- <picture>
<source
media="(prefers-color-scheme: dark)"
srcset="
https://api.star-history.com/svg?repos=OpenBMB/MiniCPM-o&type=Date&theme=dark
"
/>
<source
media="(prefers-color-scheme: light)"
srcset="
https://api.star-history.com/svg?repos=OpenBMB/MiniCPM-o&type=Date
"
/>
<img
alt="Star History Chart"
src="https://api.star-history.com/svg?repos=OpenBMB/MiniCPM-o&type=Date"
/>
</picture> -->
## Key Techniques and Other Multimodal Projects <!-- omit in toc -->
👏 Welcome to explore key techniques of MiniCPM-o/V and other multimodal projects of our team:
[VisCPM](https://github.com/OpenBMB/VisCPM/tree/main) | [RLPR](https://github.com/OpenBMB/RLPR) | [RLHF-V](https://github.com/RLHF-V/RLHF-V) | [LLaVA-UHD](https://github.com/thunlp/LLaVA-UHD) | [RLAIF-V](https://github.com/RLHF-V/RLAIF-V)
## Citation <!-- omit in toc -->
If you find our model/code/paper helpful, please consider citing our papers 📝 and staring us ⭐️！
```bib
@article{yao2024minicpm,
title={MiniCPM-V: A GPT-4V Level MLLM on Your Phone},
author={Yao, Yuan and Yu, Tianyu and Zhang, Ao and Wang, Chongyi and Cui, Junbo and Zhu, Hongji and Cai, Tianchi and Li, Haoyu and Zhao, Weilin and He, Zhihui and others},
journal={arXiv preprint arXiv:2408.01800},
year={2024}
}
```
Reference in New Issue View Git Blame Copy Permalink