Atlas

Benchmarks

← All benchmarks

FrontierCode 1.0 Main - Pass Rate

Code · 2026-06-08

FrontierCode 1.0 Main pass rate on the 100 hardest tasks in the 150-task Extended set, including all 50 Diamond tasks. A trial passes only when it satisfies every blocker criterion, approximating whether a repository maintainer would merge the patch.

Top models (higher is better)

ModelScore
Claude Fable 549.3
Opus 4.837.3
GPT-5.528.3
Opus 4.726.2
Kimi K2.7 Code24.3
GLM-5.221.2
GPT-5.4 Mini20.0
Gemini 3.1 Pro Preview18.7
Kimi K2.618.0
Sonnet 4.616.9
Qwen3.6 Plus (2026-04-02)10.8
Kimi K2.57.5
MiniMax M2.76.7
SWE-1.66.1
MiniMax M2.56.0
Gemini 3.1 Flash-Lite Preview5.3
Loading Atlas data…