Laguna S 2.1

Laguna S 2.1, a 118-billion parameter Mixture-of-Experts model developed in under nine weeks, delivers strong performance on coding benchmarks despite being significantly smaller than competing models, supporting up to 1 million token context windows with extended reasoning capabilities. The model demonstrates competitive results across multiple software engineering benchmarks, positioning it as the most capable agentic coding model in its weight class, with the developers releasing full evaluation trajectories for transparency.