Poolside Ships Laguna S 2.1, an Open-Weight Coding Model That Beats Rivals 10x Its Size

My take: Poolside launched Laguna S 2.1, a 118-billion-parameter open-weight model trained from scratch in under nine weeks, now available for free on Hugging Face. On SWE-Bench Multilingual (one of the toughest coding benchmarks), it scores 78.5%, outperforming much larger models like DeepSeek-V4-Flash.

For developers building with AI, this matters for two reasons: first, access to a high-capability model with no API costs or commercial use restrictions; second, the training speed (nine weeks from scratch to launch) reflects how fast the industry is moving today. It can run on a single DGX Spark GPU at 4-bit quantization, putting it within reach of smaller teams.

If you are evaluating whether to add open-weight models to your development stack, this one is worth keeping on the short list.