MiniCPM-V 4.6
A 1.3B pocket VLM that runs on your phone — Apache 2.0, image and video understanding.
About
MiniCPM-V 4.6 is OpenBMB's pocket-sized vision-language model, released May 11, 2026: 1.3B parameters pairing a SigLIP2-400M vision encoder with a Qwen3.5-0.8B language backbone, under Apache 2.0. The pitch is efficiency without embarrassment — OpenBMB reports a 13 on the Artificial Analysis Intelligence Index (vendor-reported), the highest score for any open model under 2B parameters at the time, with chain-of-thought visual reasoning built in.
The practical story is deployment: it runs on iOS, Android, and HarmonyOS with the edge-adaptation code open-sourced, plus llama.cpp/GGUF, Ollama, and vLLM support on desktop, and OpenBMB ships a public free API key for the hosted service. A 4.7 generation followed in October 2026, but 4.6 remains the documented, ecosystem-supported reference point.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Vendor-reported: Artificial Analysis Intelligence Index 13 (highest open model under 2B parameters at release).
Independent: no third-party reproduction found.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Repository verified live on Hugging Face (HTTP 200); Apache 2.0; vLLM and llama.cpp support confirm real ecosystem adoption. Docked for: (1) the headline Intelligence Index score is vendor-reported; (2) the newer 4.7 generation is already out, so this is the previous — albeit better-documented — release.