BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

GPQA Diamond (Vals)

Vals AI's own GPQA Diamond run. Shown for reference.

As of 24 Oct 2026, Gemini 3.1 Pro leads GPQA Diamond (Vals) on BenchLeader with 95.5%, ahead of GPT-5.6 Sol at 95.2%, across 130 model configurations with a published result.

Published by
Vals AI
Category
Reasoning
Index weight
Reference only
Models
130
Data as of
24 Oct 2026
Newest result
1 Sept 2026

Vals AI (vals.ai).

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 on the 198 questions.

What to keep in mind

Vals AI's own run of GPQA Diamond. Shown for reference; Epoch AI's run is counted in the index. See also the GPQA Diamond board.

Top 15
  1. 1Gemini 3.1 Pro (high)95.5%
  2. 2GPT-5.6 Sol (max)95.2%
  3. 3Grok 4.6 (high)94.7%
  4. 4Gemini 3.8 Flash (high)94.4%
  5. 5Gemini 3.7 Flash (high)93.9%
  6. 6Qwen3.8 Max (max)93.7%
  7. 7Claude Fable 5.193.4%
  8. 8Claude Opus 593.4%
  9. 9Gemini 3.6 Flash (high)93.4%
  10. 10Claude Fable 593.2%
  11. 11GPT-5.5 (xhigh)93.2%
  12. 12Grok 4.5 (high)92.9%
  13. 13Kimi K392.9%
  14. 14Gemini 3.5 Flash (high)92.7%
  15. 15MiniMax M392.7%
130 of 130
#
1Gemini 3.1 Prohigh95.5%60.32026-09-01
2GPT-5.6 Solmax95.2%68.02026-09-01
3Grok 4.6high94.7%62.92026-09-01
4Gemini 3.8 Flashhigh94.4%64.42026-09-01
5Gemini 3.7 Flashhigh93.9%63.52026-09-01
6Qwen3.8 Maxmax93.7%63.32026-09-01
7Claude Fable 5.193.4%66.42026-09-01
8Claude Opus 593.4%65.52026-09-01
9Gemini 3.6 Flashhigh93.4%62.02026-09-01
10Claude Fable 593.2%67.12026-09-01
11GPT-5.5xhigh93.2%66.12026-09-01
12Grok 4.5high92.9%59.32026-09-01
13Kimi K392.9%59.12026-09-01
14Gemini 3.5 Flashhigh92.7%63.12026-09-01
15MiniMax M392.7%56.42026-09-01
16DeepSeek V4 Promax92.4%60.82026-09-01
17Claude Opus 4.892.4%60.12026-09-01
18GPT-5.4xhigh91.7%64.72026-09-01
19GPT-5.2xhigh91.7%60.72026-09-01
20Gemini 3 Prohigh91.7%60.02026-09-01
21GPT-5.6 Lunamax91.7%59.92026-09-01
22Grok 4.391.4%54.72026-09-01
23Muse Spark 1.1xhigh91.2%61.62026-09-01
24GPT-5.6 Terraxhigh90.9%62.82026-09-01
25Claude Opus 4.790.2%63.32026-09-01
26Qwen3.7max90.2%60.62026-09-01
27DeepSeek V4 Flashhigh89.9%57.42026-09-01
28Muse Spark89.7%64.32026-09-01
29Claude Opus 4.6thinking89.7%62.52026-09-01
30Kimi K2.689.1%55.42026-09-01
31Qwen3.8 27Bxhigh88.9%56.22026-09-01
32Claude Sonnet 588.9%56.12026-09-01
33Grok 4.20thinking88.6%58.62026-09-01
34GLM 5.3max88.1%63.52026-09-01
35Grok 488.1%58.02026-09-01
36Gemini 3 Flashhigh87.9%58.32026-09-01
37Qwen3.6 Plus87.4%58.82026-09-01
38Qwen3.5 Plusthinking87.4%57.32026-09-01
39GPT-5.1high86.6%57.72026-09-01
40MiniMax M2.786.6%55.62026-09-01
41GLM 5.3 Flashmax86.4%54.22026-09-01
42Nemotron 3 Ultra 550B A55B86.1%55.52026-09-01
43Claude Opus 4.5thinking85.9%59.92026-09-01
44GPT-5high85.6%57.92026-09-01
45Claude Sonnet 4.685.6%57.12026-09-01
46GLM 5.285.6%52.02026-09-01
47Grok 4 Fastthinking85.3%51.42026-09-01
48Qwen3 Maxmax84.8%52.82026-09-01
49Ling 3.0 Flash84.8%52.82026-09-01
50GLM 5.184.5%55.92026-09-01
51Grok 4.1 Fastthinking84.3%54.72026-09-01
52Kimi K2.5thinking84.1%57.62026-09-01
53o3high84.1%53.92026-09-01
54Gemini 3.5 Flash Litehigh83.8%47.92026-09-01
55Inkling Small83.6%54.82026-09-01
56GLM-5thinking83.3%58.92026-09-01
57GPT-5.4 Minixhigh83.1%56.32026-09-01
58Qwen3.5 Flash82.8%52.02026-09-01
59MiMo-V2.5-Pro82.6%54.92026-09-01
60MiniMax M2.582.1%53.62026-09-01

More reasoning benchmarks · All LLM benchmarks

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

GPQA Diamond (Vals): questions

What does GPQA Diamond (Vals) 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 (Vals)?
Gemini 3.1 Pro leads GPQA Diamond (Vals) with 95.5% as of 24 Oct 2026, ahead of GPT-5.6 Sol at 95.2%.
How many models have GPQA Diamond (Vals) results?
130 model configurations have a GPQA Diamond (Vals) result on BenchLeader, all taken from Vals AI.
Who runs GPQA Diamond (Vals) and how often is it updated?
GPQA Diamond (Vals) is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does GPQA Diamond (Vals) count toward the BenchLeader Index?
No. GPQA Diamond (Vals) is shown for reference but left out of the composite index.