Cognition Launches SWE-2 Flash for Faster, Lower-Cost Coding Agents

Cognition launched SWE-2 Flash, a coding AI model with 276B parameters, running three times faster and costing one-fifth as much as SWE-2. It targets routine software-agent tasks and offers free use through November 30, 2026. Benchmarks show it outperforms similar-sized models but trades some accuracy for speed and cost efficiency.