Atlas

Benchmarks

← All benchmarks

FrontierCode 1.1 Main - Pass Rate

Code · 2026-07-07

FrontierCode 1.1 Main pass rate on the 100 hardest tasks in the 150-task Extended set. A trial passes only when it satisfies every blocker criterion and is not flagged for unfair internet use. The result averages repeated trials per task.

Top models (higher is better)

ModelScore
Claude Opus 5.560.0
Claude Fable 558.9
Claude Opus 558.9
GPT-6 Astra58.8
Claude Sonnet 5.557.2
GPT-6.1 Sol55.8
Claude Fable 5.155.6
SWE-255.5
GPT-6 Sol54.3
Grok 4.653.1
Grok 4.753.1
GPT-5.6 Sol52.9
Gemini 4 Argon52.8
Claude Haiku 5.551.6
Opus 4.851.6
Gemini 3.7 Flash48.9
DeepSeek V4.1 Flash48.6
GPT-5.548.2
GPT-6 Luna47.8
SWE-1.747.8
Sonnet 547.6
Grok 4.547.2
Gemini 3.8 Flash46.7
GPT-5.6 Terra46.3
GPT-5.6 Luna44.7
GLM-5.344.7
Mistral Large 4 Preview43.5
Opus 4.742.8
Gemini 3.6 Flash38.9
GLM-5.3 Flash35.7
Kimi K2.7 Code33.6
DeepSeek V4 Pro 081331.8
GPT-5.4 Mini30.8
Opus 4.629.7
Composer 2.529.3
Sonnet 4.627.5
GLM-5.227.4
DeepSeek-V4-Flash-073121.1
DeepSeek-V4-Pro20.0
MiniMax M316.6
Loading Atlas data…