Poolside releases Laguna S 2.1, an open-weight coding model it says beats rivals many times its size

By NewSunSEO | July 26, 2026

Poolside, a San Francisco artificial intelligence lab that has spent most of its three-year existence selling coding models to governments and defense agencies, released Laguna S 2.1 on Tuesday, July 21, 2026. The 118-billion-parameter Mixture-of-Experts (MoE) model activates only 8 billion parameters per token and supports a context window of up to 1 million tokens. Weights are available immediately on Hugging Face under the permissive OpenMDW-1.1 license.

What does Laguna S 2.1 benchmark against?

Poolside reports that Laguna S 2.1 scores 70.2 percent on Terminal-Bench 2.1, which measures long-horizon terminal tasks. That result places the model 11th on the company’s compiled leaderboard and ahead of DeepSeek-V4-Pro-Max (a 1.6-trillion-parameter model) at 64.0, Thinking Machines Inkling at 63.8 (975 billion parameters), and Nvidia Nemotron 3 Ultra at 56.4 (550 billion parameters). On SWE-Bench Multilingual, Laguna S 2.1 posts 78.5 percent, and 59.4 percent on the SWE-Bench Pro public dataset.

How was the model trained?

Pre-training began on May 22, and the model launched in under nine weeks, trained on 4,096 Nvidia H200 GPUs. Poolside says it has shipped three models in three months. The sparse MoE architecture uses 256 routed experts plus one shared expert, grouped-query attention, and interleaved sliding-window layers. Because inference cost scales with the 8 billion active parameters, the company notes the model is small enough to run on a single Nvidia DGX Spark. On its hardest benchmark with thinking mode enabled, the model consumes roughly 249,000 completion tokens per trajectory.

Why is Poolside framing this as a Western open-weight release?

Poolside is positioning Laguna S 2.1 as a response to the dominance of Chinese open-weight labs, naming DeepSeek, Qwen, Kimi, GLM, MiniMax, and Tencent Hunyuan. The company also notes that Laguna S 2.1 occupies a size class into which no Western lab has released open weights in 11 months, since OpenAI’s gpt-oss-120b last August.

Co-CEO Jason Warner said: “The West needs open-weight models it can trust, run, and build on.” Co-founder and co-CEO Eiso Kant wrote on X: “I believe intelligence should and will become a commodity.”

Where can developers get the weights?

The weights are available on Hugging Face under the OpenMDW-1.1 license, which Poolside describes as permissive. The 1 million token context window, sparse activation pattern, and single-node DGX Spark footprint are the operational features the company highlights for developers evaluating the release.

FAQ

What is Laguna S 2.1?

Laguna S 2.1 is a 118-billion-parameter Mixture-of-Experts coding model from Poolside released on July 21, 2026. It activates 8 billion parameters per token, supports up to a 1 million token context window, and is available on Hugging Face under the OpenMDW-1.1 license.

How does Laguna S 2.1 perform on coding benchmarks?

Poolside reports 70.2 percent on Terminal-Bench 2.1, 78.5 percent on SWE-Bench Multilingual, and 59.4 percent on the SWE-Bench Pro public dataset.

Why does Poolside describe Laguna S 2.1 as a Western open-weight release?

Poolside frames the release as a response to the dominance of Chinese open-weight labs and notes that no Western lab has released open weights in its size class for 11 months, since OpenAI’s gpt-oss-120b in August 2025.

Related coverage

• Bessent threatens sanctions over Chinese AI models while Huang defends open-weight rivals
• Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open Model Rivaling Top Closed Systems