MiniCPM5-2B
#11 in Small & Edge ModelsOpenbmb · v5 · 2b · since 7. September 2026 · 3× · last seen Sep 10, 2026
44
Momentum
MiniCPM5-2B is a dense language model released by OpenBMB (ModelBest) on September 6/7, 2026, with about 2.52 billion parameters (1.98 billion non-embedding), optimized for edge and on-device deployment. It uses the standard LlamaForCausalLM architecture with 42 layers and grouped-query attention, supports a native 131,072-token context window, and offers hybrid Think/No-Think reasoning plus native tool calling. The model is released as open weights under the Apache-2.0 license, along with training data and RL recipes, and achieves an average score of 53.9 across a 34-benchmark comparison set, surpassing 4B-class models according to the vendor.
Momentum trend
13.06.11.09.
Features
| Key Benchmark (%) | Average 53.9 across 34 benchmarks; Artificial Analysis Intelligence Index score 23 (#1 among open-source models under 4B) |
| Context Window (Tokens) | 131,072 tokens (native) |
| License | Apache 2.0 |
| Multimodality | Text-only language model, no native image/audio multimodality documented |
| Platform | Hugging Face, GitHub, ModelScope; runs via vLLM, SGLang, Transformers, llama.cpp, Ollama, LM Studio, MLX and FlagOS |
| Release Date | September 7, 2026 (weights already pushed to Hugging Face on September 6, 2026) |