Cognition has announced the launch of SWE-1.7, its most capable model to date, which achieves frontier-level intelligence at a significantly lower cost. This advancement is the result of broad improvements across Cognition’s reinforcement learning (RL) pipeline, including better infrastructure, more stable training, higher-quality data, and new techniques for long-horizon tasks.
What Shipped
SWE-1.7 is available today in Devin via Cerebras at 1000 TPS. The model has been optimized for longer-horizon asynchronous tasks, a crucial component of high-quality software engineering. The launch of SWE-1.7 challenges the idea of a ‘post-training ceiling’ and suggests that RL can push capabilities much further than previously believed.
Implications for Builders
The introduction of SWE-1.7 has significant implications for builders, as it provides a more cost-effective solution for achieving frontier-level intelligence. This could lead to increased adoption and innovation in the field. However, the full extent of SWE-1.7’s capabilities and potential applications is still being explored.
Sources
- Cognition: SWE-1.7: Frontier Intelligence at a Fraction of the Cost *Fan coverage from Devin Central — not an official Cognition announcement. Devin is a trademark of Cognition.