GPQA Diamond (Epoch AI runs)
198 graduate-level biology, physics and chemistry questions that non-experts cannot look up. Run by Epoch AI.
GPT-6 Astra
#1 right now
95.8%
Accuracy
100
Models ranked
Sep 2, 2026
Board published
Ranking
Best variant per model, as published. Higher is better.
- 195.8%
GPT-6 Astra
as “gpt-6-astra_max” - 295.4%
Gemini 3.8 Flash
as “gemini-3.8-flash_high” - 394.8%
Gemini 3.7 Flash
as “gemini-3.7-flash_high” - 494.6%
GPT-5.4 Pro
as “gpt-5.4-pro-2026-03-05_xhigh” - 594.4%
Gemini 3.1 Pro Preview
as “gemini-3.1-pro-preview_high” - 694.1%
Gemini 3.6 Flash
as “gemini-3.6-flash_high” - 794.0%
GPT-5.5 Pre Release
as “gpt-5.5-pre-release_xhigh” - 794.0%
Grok 4.6
as “grok-4.6_high” - 993.9%
GPT-5.5 Pro Pre Release
as “gpt-5.5-pro-pre-release_xhigh” - 1093.9%
Claude Opus 5
as “claude-opus-5_max” - 1193.5%
GPT-5.6 Sol
as “gpt-5.6-sol_max” - 1293.4%
Grok 4.5
as “grok-4.5_high” - 1393.3%
GPT-5.6 Terra
as “gpt-5.6-terra_max” - 1493.3%
GPT-5.4
as “gpt-5.4-2026-03-05_xhigh” - 1593.1%
Kimi K3
as “kimi-k3_max” - 1692.8%
Gemini 3.5 Flash
as “gemini-3.5-flash_high” - 1792.7%
Qwen3.8 Max (0902)
as “qwen3.8-max_xhigh” - 1892.6%
Gemini 3 Pro
as “gemini-3-pro-preview” - 1991.9%
GLM-5.2
as “glm-5.2_max” - 2091.7%
DeepSeek V4 Pro 0813
as “deepseek-v4-pro-0813_max” - 2191.6%
GPT-5.6 Luna
as “gpt-5.6-luna_max” - 2291.4%
GPT-5.2
as “gpt-5.2-2025-12-11_xhigh” - 2391.0%
Claude Opus 4.8
as “claude-opus-4-8_max” - 2391.0%
DeepSeek V4 Flash Vision
as “deepseek-v4-flash-0731_max” - 2590.9%
GLM-5.3
as “glm-5.3_max”
Rank over time
History builds with every daily capture. One capture so far.
Other reasoning boards
AA Intelligence
#1 Claude Opus 5.5
HLE
#1 GPT-6 Astra
Epoch ECI
#1 GPT-6 Astra
ARC-AGI-2
#1 GPT-6 Astra
LiveBench
#1 Claude Fable 5.1
MMLU-Pro
#1 Gemini 3.1 Pro Preview
Open LLM LB
#1 none
Scores as published by Epoch AI on the capture date (2026-09-24). Source: epoch.ai benchmark data (CC BY).