evals.report
BenchmarksLabsCompareRun guidesIn the wild

SWE-2

Cognition · SWE. Released Sep 10, 2026.

SWE-2 is a model from Cognition in the SWE family, released Sep 10, 2026. evals.report tracks 2 reported SWE-2 benchmark scores across FrontierCode, Terminal-Bench 2.1 — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

2 results

Benchmark results 2

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierCodeCoding50.0%weighted score (Main)OfficialSep 10, 2026Details
Terminal-Bench 2.1Agents92.8%task successVerifiedSep 10, 2026Details