Atlas

Benchmarks

← All benchmarks

SUPERChem

Science

Peking University's SUPERChem: 500 expert-curated, reasoning-intensive chemistry problems available in multimodal and text-only forms. Scores are final-answer accuracy.

Top models (higher is better)

ModelScore
GPT-6 Astra80.3
Claude Opus 5.579.6
GPT-6.1 Sol77.8
GPT-5.6 Sol Pro74.9
Claude Fable 5.174.7
Claude Opus 574.7
Gemini 3.7 Flash72.8
GPT-6 Sol72.8
Gemini 3.8 Flash72.6
GPT-5.6 Sol71.0
Gemini 3.5 Flash68.3
GPT-5.566.8
Gemini 3.6 Flash66.0
Grok 4.665.7
Grok 4.565.5
GPT-5.6 Terra64.9
Claude Fable 564.4
GPT-5.464.1
Kimi K363.7
Claude Haiku 5.562.0
Opus 4.860.3
GPT-5.6 Luna58.1
DeepSeek V4.1 Flash56.7
GPT-6 Luna55.5
Sonnet 555.3
Kimi K2.7 Code51.5
Opus 4.746.9
Sonnet 4.646.3
MiniMax M341.8
Loading Atlas data…