Poolside has launched the Laguna S 2.1, an open-weight coding model presented as the Western counterpart to DeepSeek and Qwen.

      Poolside has launched Laguna S 2.1, a 118-billion-parameter open-weight coding model designed for agentic coding, which the San Francisco startup claims can match or surpass models many times its size. Utilizing a mixture-of-experts architecture with eight billion active parameters per token, this compact model can operate on a single Nvidia DGX Spark desktop system, and its weights are accessible on Hugging Face under the OpenMDW license from the Linux Foundation.

      In evaluations on Terminal-Bench and SWE-Bench Pro—two tests for agentic coding—Laguna S 2.1 achieved scores of just over 70 percent and nearly 60 percent, respectively, competing with models from DeepSeek, Nvidia, and Thinking Machines that have two to eight times more active parameters. Poolside acknowledges that the model has not yet reached the cutting edge, as closed-source models from OpenAI and Anthropic score significantly higher on the same evaluations.

      The launch is strategically positioned as a response to the prevalence of Chinese labs in the open-weight sector, where DeepSeek, Alibaba's Qwen family, and Moonshot's Kimi have led for over a year. According to the company, no Western lab had released an open-weight model in the 118-billion-parameter range for 11 months prior to this launch. Forbes mentioned that Poolside aimed to offer Western businesses and governments a self-hosted alternative to avoid sending data to foreign providers.

      Founded in 2023 by Jason Warner, a former chief technology officer at GitHub, and Eiso Kant, Poolside secured $500 million in Series B funding in October 2024, achieving a $3 billion valuation with support from Nvidia and eBay. However, a planned $2 billion Series C that would value the company at $14 billion failed to materialize when CoreWeave exited a joint data center project in Texas in April 2026. Currently, the company serves government, defense, and other highly regulated sectors via its API and agent harness.

      Poolside states it developed the model through its internal Model Factory platform, which automates architecture search and reinforcement learning from code execution, completing training in under four weeks using 4,000 Nvidia H200 GPUs. The smaller Laguna XS was released three weeks prior, and the company indicates that new models are launched approximately every five weeks. As a showcase for long-horizon reasoning, Poolside shared an example of the model independently solving a combinatorics problem that previously only the largest frontier models could address.

      The expectation is that enterprises will opt to operate a capable coding model on their own infrastructure rather than submitting prompts to a closed API, a theory contingent on Laguna S 2.1 performing in production as it does in benchmarks. Poolside’s own findings reveal the model trailing closed-source leaders by about 10 to 15 percentage points on Terminal-Bench, a difference significant for customers weighing the benefits of self-hosting.

      The ability of Poolside to bridge this gap in its next iteration, while contending with rapidly advancing Chinese open-weight models, will be crucial in determining whether the Western open-weight divide is a fleeting or enduring issue.