Atlas

Benchmarks

← All benchmarks

GPQA Diamond

General QA · 2023-11-20

The hardest GPQA subset, with graduate-level Google-proof questions in biology, physics, and chemistry that require deep domain reasoning. Scores are reported as accuracy.

Top models (higher is better)

ModelScore
Gemini 4 Argon96.4
GPT-6 Astra96.2
Gemini 3.1 Pro Preview96.0
GPT-5.595.5
Fugu Ultra v1.095.5
Grok 4.695.5
Gemini 3.7 Flash95.5
GPT-5.6 Sol94.9
Qwen3.8-Max94.9
Gemini 3.8 Flash94.9
GPT-5.6 Sol Pro94.9
Opus 4.794.2
Kimi K3.1 Pro94.2
Claude Fable 594.1
Claude Fable 5.193.9
GPT-5.5 Pro93.9
Step 5 Preview93.5
Kimi K393.4
GPT-5.2 Pro93.2
DeepSeek V4.1 Flash 101793.2
Claude Opus 592.9
Grok 4.392.9
DeepSeek V4 Pro 081392.9
Grok 4.592.9
Claude Mythos Preview92.9
GPT-5.6 Terra92.9
GPT-5.492.8
Grok 4.792.7
GPT-5.3-Codex92.6
Gemini 3.6 Flash92.4
GPT-5.6 Luna92.4
MiniMax M392.4
Interfaze 192.4
Interfaze Beta92.4
Qwen3.8 Max (0902)92.3
Opus 4.892.0
GPT-5.291.9
Qwen3.8-Flash-Next91.7
Gemini 3.5 Flash91.4
GLM-5.291.4
Loading Atlas data…