Laguna S 2.1
#14 in Open-Source LLMsPoolside Ai · v2.1 · 6× · last seen Jul 24, 2026
47
Momentum
Laguna S 2.1 is an Open Source language model (Mixture-of-Experts) developed by Poolside AI with 118 billion total parameters, of which only around 8 billion are activated per Token. The model is designed for agentic programming and long-running tasks and supports a context window of up to 1 million Token in both Reasoning and non-Reasoning modes. It is distributed under the open OpenMDW-1.1 license on Hugging Face as well as through partners such as OpenRouter, Baseten, and the Poolside API, and achieved 40.4% on the DeepSWE Benchmark and 70.2% on Terminal-Bench 2.1 according to the official blog.
Momentum trend
26.04.25.07.
Features
| Key Benchmark (%) | 40.4% on DeepSWE; 70.2% on Terminal-Bench 2.1 |
| Context Window (Tokens) | up to 1,000,000 tokens (thinking and no-thinking modes) |
| License | OpenMDW-1.1 (open model license, commercial use permitted) |
| Multimodality | Text-based (no multimodal input per model specification) |
| Platform | Open-weight model, available on Hugging Face, via Poolside API, OpenRouter, Baseten, Vercel AI Gateway; local inference via vLLM, SGLang, Ollama, llama.cpp |
| Price per 1M Tokens | $0.10 input / $0.20 output per 1M tokens (1M context, via OpenRouter); free endpoint with 262K context available |
| Release Date | July 21, 2026 |