Research
Cognition trains models optimized for software engineering. The challenge is not only improving raw capability, but also shaping agent behavior. Agents should feel fast, simple, and easy to course-correct. They should respond clearly, recover gracefully from mistakes, and remain steerable as tasks become more complex. Our focus is building models that are reliable in real workflows, not just impressive in isolated benchmarks.
Articles
Introducing SWE-2 Flash
We’re introducing SWE-2 Flash, a fully American coding model that sets a new bar for intelligence at low cost and high speed.
FrontierCode 2.0
FrontierCode 2.0 is a new generation of our FrontierCode benchmark: more diverse, challenging tasks from more repositories, graded by open-source maintainers on whether agent output is actually mergeable.
Introducing SWE-2: Pushing the Pareto Frontier
Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.
Factoring RSA-260
How Devin and a Cognition researcher built the world’s highest-performance GPU lattice siever, to make factoring numbers 10x cheaper than the previous state of the art
Join the Research Team
We’re a small team working on the hardest problems in AI. You’ll work with some of the best researchers and engineers in the field, pushing models to write better code, reason better, and work more autonomously. If this excites you, come build with us.