MiniCPM5-1B

First release of the MiniCPM5 series — jumps past MiniCPM4 with a 1.08B dense Llama-architecture model featuring 128K context and hybrid thinking mode. Self-reported 1B-class open-source SOTA on agent tool-use, code, and hard reasoning vs LFM2.5-1.2B and Qwen3-0.6B / Qwen3.5-0.8B.

Shipped with SFT, MLX, and GGUF variants for on-device deployment.