Poolside releases Laguna S 2.1, 118B open-weight model beating DeepSeek V4 Flash on agentic tasks
<cite index="15-2">Poolside released Laguna S 2.1, a 118-billion-parameter open-weight foundation model built for agentic coding on July 21, 2026. On Terminal-Bench 2.0 and SWE-Bench Pro, Laguna S 2.1 matches or exceeds models several times its size, including DeepSeek-V4-Flash, NVIDIA's Nemotron 3 Ultra or Thinking Machines' Inkling. The weights are available on Hugging Face under an OpenMDW-1.1 license</cite>.
<cite index="11-3,11-4">On Token-level pricing, Laguna S 2.1 costs $0.10 input / $0.20 output per 1M tokens versus $0.14 / $0.28 for DeepSeek V4 Flash (Max), making it roughly 30% cheaper. On benchmark splits, DeepSeek V4 Flash has the edge for coding (68.8 vs 59.4 average), while Laguna S 2.1 has the edge for agentic tasks (70.2 vs 63.8)</cite>. <cite index="16-1">On SWE-Bench Multilingual, Laguna hits 78.5%, placing it in range with Tencent Hy3 and DeepSeek-V4-Pro Max, and on Toolathlon Verified it scores 49.7%, ahead of Nemotron 3 Ultra and DeepSeek-V4-Flash Max</cite>.
<cite index="15-1">The release arrives as the question of who supplies the West's open-weight models has moved from research circles into boardrooms and Washington, with developer usage shifting sharply toward open-weight systems teams can download, inspect, and run on their own infrastructure</cite>. For architects evaluating agentic coding stacks, Laguna S 2.1's local-deployable weight, cost profile, and agentic-task lead make it a practical alternative to API-only models; the trade-off is modest latency and narrower coverage outside code-specific evals.