MiniCPM5-2B lands on Hugging Face as author claims Llama architecture beats Qwen3.5

solyarisoftware · x · 2026-09-12

openbmb has released MiniCPM5-2B on Hugging Face. Benjamin Marie's takeaway: based on this model, the Llama architecture can outperform Qwen3.5, suggesting architecture choice still matters for small models. The model card adopts a Qwen-style chat template with tool-calling support.

Original post →

More from Models

Models channel →