Nanbeige 3 Billion Parameter Model

A developer testing Nanbeige 4.2, a 3-billion-parameter looped-transformer model from a Chinese company, found it well-suited for tool calling with a Rust MCP server, outperforming Gemma 4 12B in that task, though it is slower than other 3B models and has a parsing issue that drops some tool calls.

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.