OpenBMB also releases a 1B version of MiniCPM-V 4.7 with GGUF files, an on-device vision-language model (per community posts)
Verified (fact)
Following the previously covered 35B-A3B MoE version, a 1B version of MiniCPM-V 4.7 has been released on Hugging Face along with GGUF files. The 1B version also inherits the series' single-image, multi-image and video understanding capabilities. It emphasizes efficient visual encoding and on-device inference.
Needs fact-check (claim)
It claims to achieve a new Pareto frontier of performance and efficiency (specific benchmark conditions are not in the input materials).
Why it matters (interpretation)
With the same model generation available from a small 1B version up to MoE, users can pick a local vision model to match their device's specs.
Confidence medium · official source pending
Sources
- Community openbmb/MiniCPM-V-4.7-1B Reddit
Story thread
- 2026-10-12 OpenBMB releases open weights for MiniCPM-V 4.7 35B-A3B, the series' first MoE vision-language model (community reports)
- 2026-10-13 OpenBMB also releases a 1B version of MiniCPM-V 4.7 with GGUF files, an on-device vision-language model (per community posts)
More about OpenBMB
-
OpenBMB releases open weights for MiniCPM-V 4.7 35B-A3B, the series' first MoE vision-language model (community reports)
OpenBMB has released MiniCPM-V 4.7 on Hugging Face. It is an MoE model with 35B total parameters, of which about 3B are active per...