SWE-1.5
Released · cognition.com
Benchmark results
| Benchmark | Score |
|---|---|
| SWE-bench ProPublic set (731 tasks) · Cascade harness | 40.08%Corroborated The same number is printed in 2 other lab documents: 1 2 |
| SWE-bench ProSet not stated · SWE-agent harness | 34.47%Checked A person checked this number against its source on 7 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced. |
| Claude Code harness | 29.00%Checked A person checked this number against its source on 7 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced. |
Sources
Reported by Cognition
- Introducing SWE-1.5: Our Fast Agent Modelcognition.com · 29 Oct 2025
Original image 
Original image
