Poolside's Laguna S 2.1 tops open-weight agentic coding class
Poolside released Laguna S 2.1, a 118B MoE coding model with 1M context under an open license, hitting 78.5% on SWE-Bench Multilingual with only 8B active params.
Poolside released Laguna S 2.1 on July 21, an open-weight agentic coding model that the company claims is "the most capable in its weight class by a wide margin." The 118-billion-parameter Mixture-of-Experts model activates just 8 billion parameters per token, supports a 1-million-token context window, and went from training start to launch in under nine weeks.
Small activation, big scores
The benchmark numbers back the positioning. Laguna S 2.1 posts 78.5% on SWE-Bench Multilingual, 59.4% on SWE-Bench Pro, 70.2% on Terminal-Bench 2.1 — ranking 11th on that leaderboard despite being far smaller than most models above it — and 40.4% on DeepSWE v1.1. Poolside emphasizes long-horizon behaviors over raw scale: persistence on hard tasks, self-verification and recovery from dead ends, traits that matter more in autonomous coding agents than single-shot accuracy.
Distribution is aggressively open. Weights ship on Hugging Face under the Linux Foundation's OpenMDW-1.1 license in BF16, FP8, INT4 and NVFP4 formats. The model is live on Baseten, Vercel AI Gateway and OpenRouter — where it is priced at $0.10 per million input tokens and $0.20 per million output tokens, a fraction of frontier-model rates — and plugs into coding agents including Cline and Poolside's own tooling. A free, no-login web version runs at chat.poolside.ai.
The release caps a rapid cadence: three model releases in three months, following M.1 and XS.2 in April and XS 2.1 earlier in July.
Why it matters
The center of gravity in open-weight coding models keeps shifting toward small-activation MoE designs that run cheaply at scale. Laguna S 2.1 undercuts frontier coding APIs on price by an order of magnitude while landing within striking distance on agentic benchmarks — exactly the trade that matters for the swarm-style, many-agent workflows now emerging. For Poolside, long viewed as a lab overshadowed by bigger rivals, it is a credibility milestone; for the market, more evidence that competent coding intelligence is rapidly becoming a commodity.