OpenBMB Ships MiniCPM5-2B, a 2.5B Apache-2.0 On-Device Model Averaging 53.9 Across 34 Benchmarks

OpenBMB released MiniCPM5-2B on September 7 under Apache 2.0, a 2.52B-parameter dense LLM (1.98B non-embedding) with 131,072-token native context and hybrid-thinking mode. The model averages 53.9 across 34 benchmarks, beating Qwen3.5-4B at 51.1 and topping the sub-4B open leaderboard, with standout scores on tool use (97.1 on τ²-Bench Telecom) and code (69.1 on LiveCodeBench v6). Compatible with vLLM, SGLang, llama.cpp, Ollama, LM Studio, and MLX; OpenBMB also published pre-training, SFT, and RL datasets alongside intermediate checkpoints.