[KongchangAI]
BenchmarkFrontier SWE 2

Frontier SWE v2

A benchmark targeting difficult software engineering problems, used to evaluate the ability of AI models to solve complex engineering tasks.

Timeline (last 90 days)

Nov 10

On the Frontier code benchmark, Fable 5.2 improved from Opus 5's 45.5% to 49.1%; on Frontier SWE v2 it improved by about 7 percentage points.

Unverified50%
Oct 18

On the hard engineering problems in Frontier SWE v2, Astra scores 65.5 and Argon scores 55, a clear gap; Astra also leads in the scientific terminal tests.

Expired50%

All Facts (1)

Source Articles