Laguna M.1
#107 in Frontier-Spraakmodellepoolside · siet 28. April 2026 · 10× · tolest 30. Juni 2026
12
Momentum
Laguna M.1 is a proprietary Mixture-of-Experts language model from poolside, unveiled on April 28, 2026, with 225 billion total parameters and 23 billion parameters activated per token. It was trained from scratch on more than 30 trillion tokens using 6,144 NVIDIA Hopper GPUs and is specialized for agentic coding, multi-step tool use, and long-horizon software engineering tasks. The model is text-only (no vision input), supports a 256K token context window, and is available via poolside's API, OpenRouter, and Hugging Face (weights).
Momentum-Verloop
04.04.03.07.
Features
| Key Benchmark (%) | 72.5% on SWE-bench Verified; 46.9% on SWE-Bench Pro |
| Context Window (Tokens) | 256K (262K according to OpenRouter) |
| License | Proprietary/Closed-weight (weights available on request for small teams); not Apache 2.0 (unlike Laguna XS.2) |
| Multimodality | No – text only (no vision input) |
| Platform | poolside API, OpenRouter, Hugging Face, vLLM/SGLang/TensorRT-LLM |
| Price per 1M Tokens | $0.20 input / $0.40 output per 1M tokens (currently free for a limited time) |
| Release Date | April 28, 2026 |