SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence
Summary
Cognition AI Labs announces SWE-1.7, their most capable model to date, reaching frontier-level intelligence at a lower cost. The post outlines improvements in RL training, data quality, and long-horizon task handling, including self-compaction and an alternating length penalty. It also highlights multi-cluster training across three continents, fault-tolerant deployment, and benchmark results against competing models.