MiniCPM (OpenBMB / ModelBest 面壁智能)
Small on-device LLMs and multimodal models that outrun bigger rivals
Overview
MiniCPM is OpenBMB/ModelBest's (面壁智能) line of small open-weight language and multimodal models — MiniCPM-V for vision, MiniCPM-o for real-time omnimodal streaming — tuned to run on phones and edge hardware, developed out of Tsinghua's NLP lab.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 30 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with MiniCPM (OpenBMB / ModelBest 面壁智能)
- MiniCPM-V vision-language models for on-device image/video understanding
- MiniCPM-o omnimodal models with real-time streaming audio/video
- MiniCPM5-1B and MiniCPM5-2B dense on-device LLMs
- 96x video-token compression in MiniCPM-V 4.5
- Apache 2.0 code and model weights, with optional registration
- OpenAI-compatible API on ModelBest's platform (since May 2026)
- Official Ollama library support for MiniCPM-V 4.6
MiniCPM's whole argument is punching above its weight class: by ModelBest's own model cards, MiniCPM-o 4.5 (9B parameters) scores 77.6 on OpenCompass's eight-benchmark suite (78.2 in thinking mode), ahead of GPT-4o, and MiniCPM-V 4.6 beats Qwen3.5-0.8B's intelligence score while using 19x fewer tokens — genuinely useful if you're deploying vision-language models on hardware that can't run a 70B model. The newest text models, MiniCPM5-1B and 2B, target the same on-device niche. Current code and weights are Apache 2.0, with optional registration, and since May 2026 there is a first-party OpenAI-compatible API, so it is no longer self-host-only. A China Telecom-led round in February 2026 and a Rockchip deal for in-car AI Box products show it moving from research release to embedded product. Treat the benchmark lead as vendor-reported, and test on your own hardware.
What we found
MiniCPM comes out of THUNLP — Tsinghua University's NLP lab — with ModelBest (面壁智能) as the commercial arm, incubated from the lab in August 2022; Liu Zhiyuan is chief scientist and Li Dahai chairman and chief executive. The pitch has stayed consistent: match or beat much larger models on a fraction of the parameters, small enough to run on a phone. The newest release is MiniCPM5-2B, a dense 2.5-billion-parameter model for local deployment published in September 2026, following MiniCPM5-1B in May.
The multimodal line carries the headline numbers, all from ModelBest's own model cards. MiniCPM-o 4.5, a 9B end-to-end omnimodal model, averages 77.6 across OpenCompass's eight-benchmark suite in its standard mode and 78.2 in thinking mode — ahead of GPT-4o's 75.4 on that measure. MiniCPM-V 4.5 can compress six 448x448 video frames into 64 tokens, a 96x reduction, which is why it keeps appearing in edge-deployment shortlists. MiniCPM-V 4.6, built on SigLIP2-400M and Qwen3.5-0.8B for 1.3B parameters in all, scores 13 on the Artificial Analysis Intelligence Index against Qwen3.5-0.8B's 10, using 19x fewer tokens; it joined Ollama's official library in June 2026.
Licensing is simpler than it used to be. Current code and weights are released under Apache 2.0, and the registration questionnaire is optional, so the old community-licence caps no longer apply to these models. There is now a first-party route too: an OpenAI-compatible API on ModelBest's platform, launched in May 2026 for the MiniCPM-V and MiniCPM-o models, though its price list did not load for us. Otherwise you pull weights from Hugging Face or GitHub and self-host.
Money has followed the technical story. ModelBest closed a round of several hundred million yuan in February 2026 led by China Telecom, and Chinese press reports put its first-half 2026 fundraising above ¥5 billion at a valuation over ¥20 billion. In April it agreed with the chipmaker Rockchip to develop in-car AI Box products and next-generation cockpit solutions based on MiniCPM-o 4.5.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including MiniCPM (OpenBMB / ModelBest 面壁智能)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with MiniCPM (OpenBMB / ModelBest 面壁智能) unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to MiniCPM (OpenBMB / ModelBest 面壁智能) directly →
Spotted something out of date? Suggest an update →
Related reading
Alternatives to MiniCPM (OpenBMB / ModelBest 面壁智能)
More in LLMs & Chat
