BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

GPQA Diamond

Graduate-level science questions written to be Google-proof. Run by Epoch AI.

As of 24 Oct 2026, Gemini 4 Argon leads GPQA Diamond on BenchLeader with 96.4%, ahead of GPT-6 Astra at 95.8%, across 293 model configurations with a published result.

Published by
Epoch AI Benchmarking Hub
Category
Reasoning
Index weight
1.0
Models
293
Data as of
24 Oct 2026
Newest result
17 Oct 2026

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

What the test looks like

GPQA Diamond is a set of 198 multiple-choice questions in biology, physics and chemistry written by PhD-level domain experts. The questions are designed to be "Google-proof": a skilled non-expert with unlimited web access still answers only about a third correctly, while experts in the field score around two-thirds. The Diamond subset keeps only the questions that experts answered correctly and non-experts got wrong.

How it is scored

Accuracy, the share of questions answered correctly, with four options per question so random guessing scores 25%. Epoch AI runs every model with the same prompt and reports the mean over several attempts.

What to keep in mind

Top models now score above 85%, so the benchmark is close to saturated and small gaps are within noise: a two-point difference is about four questions. Because it is multiple choice, it rewards recognising the right answer rather than deriving it.

Top 15
  1. 1Gemini 4 Argon (high)96.4%
  2. 2GPT-6 Astra (max)95.8%
  3. 3Claude Sonnet 5.5 (max)95.6%
  4. 4GPT-6.1 Sol (max)95.4%
  5. 5Gemini 3.8 Flash (high)95.4%
  6. 6Gemini 3.7 Flash (high)94.8%
  7. 7GPT-5.4 Pro (xhigh)94.6%
  8. 8Gemini 3.1 Pro (high)94.4%
  9. 9GPT-6 Sol (max)94.3%
  10. 10Gemini 3.6 Flash (high)94.1%
  11. 11Gemini 3.1 Pro94.1%
  12. 12Grok 4.6 (high)94.0%
  13. 13GPT-5.5 Pre Release (xhigh)94.0%
  14. 14GPT-5.5 Pro Pre Release (xhigh)93.9%
  15. 15Claude Opus 5 (max)93.9%
293 of 293
#
1Gemini 4 Argonhigh96.4%70.82026-09-30
2GPT-6 Astramax95.8%70.42026-09-03
3Claude Sonnet 5.5max95.6%69.32026-09-28
4GPT-6.1 Solmax95.4%69.92026-09-29
5Gemini 3.8 Flashhigh95.4%64.42026-09-02
6Gemini 3.7 Flashhigh94.8%63.52026-08-13
7GPT-5.4 Proxhigh94.6%61.92026-03-05
8Gemini 3.1 Prohigh94.4%60.32026-02-19
9GPT-6 Solmax94.3%66.42026-09-22
10Gemini 3.6 Flashhigh94.1%62.02026-07-21
11Gemini 3.1 Pro94.1%60.62026-02-19
12Grok 4.6high94.0%62.92026-08-12
13GPT-5.5 Pre Releasexhigh94.0%–2026-04-23
14GPT-5.5 Pro Pre Releasexhigh93.9%–2026-04-23
15Claude Opus 5max93.9%68.82026-07-24
16GPT-5.6 Solmax93.5%68.02026-07-09
17Grok 4.5high93.4%59.32026-07-08
18GPT-5.6 Terramax93.3%64.42026-07-09
19GPT-5.4xhigh93.3%64.72026-03-05
20NewDeepSeek V4.1 Flash 1017max93.2%64.02026-10-17
21Grok 4.6xhigh93.2%62.52026-08-12
22Kimi K3max93.1%64.72026-07-16
23Claude Opus 592.9%65.52026-07-24
24Gemini 3.5 Flashhigh92.8%63.12026-05-19
25Grok 4.7xhigh92.7%61.62026-09-21
26Qwen3.8 Maxxhigh92.7%57.12026-08-02
27Gemini 3 Pro92.6%60.42025-11-18
28Kimi K3high91.9%–2026-07-16
29GLM 5.2max91.9%62.32026-06-16
30DeepSeek V4 Promax91.7%60.82026-08-13
31GPT-5.6 Lunamax91.6%59.92026-07-09
32GPT-5.2xhigh91.4%60.72025-12-11
33Claude Opus 4.8max91.0%63.32026-05-28
34DeepSeek V4 Flashmax91.0%59.42026-07-31
35GLM 5.3max90.9%63.52026-08-14
36Qwen3.7max90.9%60.62026-05-19
37DeepSeek V4 Prohigh90.9%57.52026-04-24
38MiniMax M390.9%56.42026-06-01
39Kimi K2.690.8%55.42026-04-20
40GPT-5.5low90.7%59.72026-04-23
41Claude Opus 5.5max90.6%72.62026-09-22
42Claude Opus 4.690.5%62.62026-02-05
43Claude Sonnet 5xhigh90.5%59.02026-06-30
44GPT-6 Lunamax90.5%60.02026-09-22
45Claude Opus 4.7xhigh90.2%57.42026-04-16
46GLM 5.3 Flashmax90.2%54.22026-08-20
47GPT-5.6 Sollow89.9%62.22026-07-09
48GPT-5.4high89.9%57.92026-03-05
49GLM 5.189.9%55.92026-04-07
50DeepSeek V4.1 Flashmax89.8%61.22026-09-09
51Muse Spark89.8%64.32026-04-08
52Claude Haiku 5.5max89.7%60.52026-10-07
53Gemini 3 Flashhigh89.4%58.32025-12-17
54Grok 4.20thinking89.3%58.62026-02-17
55GPT-5.4medium88.9%55.02026-03-05
56Gemini 3.5 Flashlow88.9%–2026-05-19
57Grok 4.3high88.8%57.02026-04-17
58NewMuse Spark Open 816B A42Bxhigh88.6%58.62026-10-15
59Inkling Smallxhigh88.5%47.62026-07-15
60Qwen3.6 Plus88.4%58.82026-03-31

More reasoning benchmarks · All LLM benchmarks

Cite as: BenchLeader, “GPQA Diamond leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/gpqa_diamond.

GPQA Diamond: questions

What does GPQA Diamond measure?
GPQA Diamond is a set of 198 multiple-choice questions in biology, physics and chemistry written by PhD-level domain experts. The questions are designed to be "Google-proof": a skilled non-expert with unlimited web access still answers only about a third correctly, while experts in the field score around two-thirds. The Diamond subset keeps only the questions that experts answered correctly and non-experts got wrong. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads GPQA Diamond?
Gemini 4 Argon leads GPQA Diamond with 96.4% as of 24 Oct 2026, ahead of GPT-6 Astra at 95.8%.
How many models have GPQA Diamond results?
293 model configurations have a GPQA Diamond result on BenchLeader, all taken from Epoch AI Benchmarking Hub.
Who runs GPQA Diamond and how often is it updated?
GPQA Diamond is published by Epoch AI Benchmarking Hub. BenchLeader re-reads the published results once a day and records the date each result was published.
Does GPQA Diamond count toward the BenchLeader Index?
Yes. GPQA Diamond contributes to the reasoning category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.