Now that I have written my Rust MCP server, I need a local agent to use it to call tools effectively. I found that various models including Gemma 4, Qwen 3.6 and Bonsai 27B can do this, but I tried a new model release called Nanbeige 4.2. This is a model from a Chinese company that seems to outperform other models of similar sizes.
Nanbeige is a Chinese word that means "bridging north and south" or something like that—it has no relation to the color beige. It uses a looped transformer, meaning it repeats the layers in the model, which means it may have closer to 6 billion parameters but they are compressed to 3 billion. It seems to be slower than other 3B models, and closer in speed to a 9B model.
For tool calling, I found that: • Nanbeige is indeed well suited for calling tools—the claims it outperforms Gemma 4 12B at this may be justified.
• The model seems able to call my MCP and built-in llama-cpp
tools when needed.
• There is a parsing issue that means some tool calls are dropped.
Overall I was impressed with Nanbeige; it is not clear whether I will switch to it as my primary non-coding model, but it has potential. It is worth keeping an eye on future Nanbeige releases.