I am continually interested in having a small, local LLM (running in llama-cpp
) that can edit and add simple features to programs written in Rust and similar languages. Recently support for the Laguna models from Poolside AI (an American AI company) was added to llama-cpp
. I tried out the smaller version (Laguna XS) locally.
I used OpenCode and had the model perform a series of refactoring changes on the code base. It made relatively few errors and almost all of the tool calls worked correctly (it became confused with escaped quotes once, and I don't blame it).
I used the official GGUF 4-bit quantized model from Poolside on their Hugging Face page, Laguna-XS-2.1-Q4_K_M.gguf
. It weighed in at 20.3 GB (good thing I have fairly fast Internet these days).
Overall, I found Laguna XS to perform well, probably better than Qwen 3.6 35B, Gemma 4 and North Mini Code. I started feeling more confident in it, and made a series of refactoring changes which mostly worked correctly. It struggled on an open-ended task ("optimize this function") but for more targeted changes, I feel Laguna XS is likely better than the other open-weights models supported by my 12 GB Nvidia GPU.