面壁智能 MiniCPM-V 4.6 下载量达200万。该小尺寸多模态模型在 Artificial Analysis Index 得分13，超过 Qwen3.5-0.8B（10分）及其思考变体（11分），token 消耗少19倍/43倍；TTFT 75.7ms，比 Qwen3.5-0.8B 快2.2倍（3136²高分辨率）。原生支持 iOS、Android、HarmonyOS，边缘代码开源；支持 vLLM、SGLang、llama.cpp、Ollama 部署及 GGUF/BNB/AWQ/GPTQ 量化。
🚀 MiniCPM-V 4.6 has just hit 2,000,000 downloads!
The pocket-sized MLLM that delivers big performance with tiny resources.
Thank you to the entire community for making this possible! ❤️
🔥 Scores 13 on Artificial Analysis Index – beating Qwen3.5-0.8B (10) & Thinking variant (11) with 19x/43x fewer tokens.
⚡ TTFT 75.7ms – 2.2x faster than Qwen3.5-0.8B, even at 3136² high-res.
📱 Native support for iOS, Android & HarmonyOS – all edge code open-sourced.
🛠️ Developer-ready – vLLM, SGLang, llama.cpp, Ollama, and GGUF/BNB/AWQ/GPTQ quantizations.
Try it:
Hugging Face: https://huggingface.co/openbmb/MiniCPM-V-4.6
GitHub: https://github.com/OpenBMB/MiniCPM-V Modelscope:https://modelscope.cn/models/OpenBMB/MiniCPM-V-4.6
来源：OpenBMB · x.com