Woofun AI reports that Cognition has deployed SWE-1.7, a large-scale programming model built on Kimi K2.7 Code, which is now available within the Devin platform. The model achieves a processing speed of 1000 tokens per second through integration with Cerebras inference chips, matching the performance of GPT-5.5 and Claude Opus 4.8 in agent programming tests while significantly lowering generation costs.

To support ultra-long tasks lasting up to 6 hours, SWE-1.7 automatically summarizes context at critical points. Training incorporates an alternating length penalty to streamline inference for simple tasks, reserving complex processing for difficult challenges.

Additionally, a multi-cluster distributed architecture across three continents enables global parameter updates in minutes by synchronizing only weight changes.