OpenBMB ships MiniCPM-V 4.7 35B: beats Qwen3.6 on 24 of 34 benchmarks
petrusenko_max · x · 2026-10-15
OpenBMB released MiniCPM-V 4.7 35B-A3B, targeting Qwen3.6 35B-A3B and filling a size gap Qwen skipped in 3.8. Apache 2.0 license, with bf16, fp8, int4, awq and gptq weights; llama.cpp support merged upstream before launch, plus a dspark draft and a 1B sibling model.
Claimed to beat Qwen3.6 35B-A3B on 24 of 34 benchmarks, including OmniDocBench (94.24 vs 80.17), CharXiv reasoning (75.6 vs 67.5), and τ²-bench (84.65 vs 79.65); independent tests underway.
More from Models
- Hacker turns EmbeddingGemma 2 into 288x realtime STT and a fast TTS — kastnerkyle · 2026-10-15
- Researcher calls out Meta for releasing a model without pricing, calling it misleading — eliebakouch · 2026-10-15
- Liquid AI's LFM2.5-24B-A2B tops IFEval with just 2.45B active parameters — huggingface · 2026-10-15
- Mollick: Gemini 4 launch shows Google following trends, not its usual product innovation — emollick · 2026-10-15
- StepFun's Step 5 Preview ships open weights, demoed porting a full game codebase to TypeScript live — Aizkmusic · 2026-10-15
- Chinese Creator Says 'Opus 5.5' Skills Can Build a Channel From Scratch — xiaohu · 2026-10-15