An American AI company has released its proprietary model 'Laguna M.1' as an open model; it is high-performance but inferior to Chinese open models.

Poolside, an American AI company, has released its proprietary model, ' Laguna M.1 ,' as an open model. Laguna M.1, originally released in April 2026 and distributed via API, is now available for anyone to download. A quantized model that can be run locally on Macs has also been released.

Today we're releasing the weights for Laguna M.1,
our most capable model to date, with a 256K context length.
Both base and post-trained checkpoints are now available on Hugging Face under Apache 2.0.

pic.twitter.com/gMWuYo8zN1

— Poolside (@poolsideai) June 18, 2026

On April 28, 2026, Poolside released its proprietary AI models, 'Laguna M.1' and 'Laguna XS.2'. Laguna M.1 is a large-scale model with 225 billion total parameters and 23 billion active parameters, while Laguna XS.2 has 33 billion total parameters and 3 billion active parameters and is a model that is relatively easy to run locally. At the time of release, only Laguna XS.2 was released as an open model, and Laguna M.1 was offered as a paid product that could be used via API. With this announcement, Laguna M.1 has also been made an open model.

American company releases high-performance, locally-running open-source model 'Laguna XS.2' - Can it compete with Chinese companies making great strides in the open-source market? - GIGAZINE

The graph below shows the benchmark results for 'Laguna M.1 (225B-A23B)', 'Devstral 2 (123B dense)', 'GLM-4.7 (355B-A32B)', 'DeepSeek-V4-Flash (284B-A13B)', 'Qwen3.5 (397B-A17B)', and 'Claude Sonnet 4.6 (number of parameters unknown)'. The Laguna M.1 beats the French-made Devstral 2, but scores lower than the Chinese-made DeepSeek-V4-Flash and Qwen 3.5.

Laguna M.1 is available at the following link. It is licensed under the Apache License 2.0 .

poolside/Laguna-M.1 · Hugging Face
https://huggingface.co/poolside/Laguna-M.1

Furthermore, several quantization models have already been released by the community. Among them, ' Laguna-M.1-MLX-Q3 ' is said to be executable locally on a Mac with an M3 Max processor and 128GB of memory.

someone asked if they could try getting Laguna M.1 running on a Mac.

We said yes.

they came back with a 3-bit MLX build running locally on Apple Silicon: ~26 tok/s, with ~100 GB peak memory on an M3 Max with 128 GB unified.

absolute GOAT behavior from @auchs …

— Poolside (@poolsideai) June 18, 2026