Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

Cognition has launched SWE-2, an advanced coding model that achieves 50.0% on FrontierCode 1.1 benchmarks while being 64% cheaper than comparable competitors like Fable 5.1, matching their performance at a fraction of the cost. The model uses a novel reinforcement learning approach scaled to the multi-trillion-parameter regime that trains multiple reasoning-effort levels simultaneously, optimizing the entire cost-performance frontier. SWE-2 is now available on Devin Desktop, CLI, Web, and Fusion platforms.