Cognition SWE-1.5 and SWE-grep Coding Agents Deployed on Cerebras Inference

Cognition's SWE-1.5 frontier coding model and SWE-grep code-search sub-agents are running on Cerebras Inference at ~1,000 tokens/second, powering Windsurf and Devin with up to 13x faster performance than competing models on GPUs.