BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

MMLU-Pro

Harder, reasoning-heavy successor to MMLU. Run by Vals AI.

As of 24 Oct 2026, Claude Fable 5.1 leads MMLU-Pro on BenchLeader with 92.4%, ahead of Claude Opus 5 at 91.6%, across 130 model configurations with a published result.

Published by
Vals AI
Category
Knowledge
Index weight
1.0
Models
130
Data as of
24 Oct 2026
Newest result
1 Sept 2026

Vals AI (vals.ai).

What the test looks like

MMLU-Pro is a harder successor to MMLU: about 12,000 questions across 14 subjects with ten answer options each and more reasoning-heavy items, removing the trivial and noisy questions of the original.

How it is scored

Accuracy, run by Vals AI.

What to keep in mind

Still multiple choice, so it tests recognition; and like all broad knowledge exams it is near saturation for frontier models.

Top 15
  1. 1Claude Fable 5.192.4%
  2. 2Claude Opus 591.6%
  3. 3Claude Fable 591.5%
  4. 4Gemini 3.1 Pro (high)91.0%
  5. 5Gemini 3.8 Flash (high)90.2%
  6. 6Gemini 3.7 Flash (high)90.1%
  7. 7Gemini 3 Pro90.1%
  8. 8Claude Opus 4.789.9%
  9. 9Claude Opus 4.889.6%
  10. 10Gemini 3.5 Flash (high)89.5%
  11. 11Grok 4.6 (high)89.4%
  12. 12Qwen3.7 (max)89.3%
  13. 13Gemini 3.6 Flash (high)89.3%
  14. 14Grok 4.5 (high)89.2%
  15. 15Claude Opus 4.6 (thinking)89.1%
130 of 130
#
1Claude Fable 5.192.4%66.42026-09-01
2Claude Opus 591.6%65.52026-09-01
3Claude Fable 591.5%67.12026-09-01
4Gemini 3.1 Prohigh91.0%60.32026-09-01
5Gemini 3.8 Flashhigh90.2%64.42026-09-01
6Gemini 3.7 Flashhigh90.1%63.52026-09-01
7Gemini 3 Pro90.1%60.42026-09-01
8Claude Opus 4.789.9%63.32026-09-01
9Claude Opus 4.889.6%60.12026-09-01
10Gemini 3.5 Flashhigh89.5%63.12026-09-01
11Grok 4.6high89.4%62.92026-09-01
12Qwen3.7max89.3%60.62026-09-01
13Gemini 3.6 Flashhigh89.3%62.02026-09-01
14Grok 4.5high89.2%59.32026-09-01
15Claude Opus 4.6thinking89.1%62.52026-09-01
16GPT-5.6 Solmax89.1%68.02026-09-01
17Muse Spark 1.1xhigh88.7%61.62026-09-01
18Qwen3.8 Maxmax88.6%63.32026-09-01
19Gemini 3 Flashhigh88.6%58.32026-09-01
20Muse Spark 1.2xhigh88.3%63.22026-09-01
21GPT-5.5xhigh88.1%66.12026-09-01
22Kimi K3max88.0%64.72026-09-01
23Claude Opus 4.1thinking87.9%56.42026-09-01
24Qwen3.6 Plus87.7%58.82026-09-01
25Kimi K2.687.6%55.42026-09-01
26Claude Sonnet 587.5%56.12026-09-01
27GPT-5.4xhigh87.5%64.72026-09-01
28Claude Sonnet 4.5thinking87.4%52.62026-09-01
29Claude Sonnet 4.687.3%57.12026-09-01
30Muse Spark87.3%64.32026-09-01
31Claude Opus 4.5thinking87.3%59.92026-09-01
32DeepSeek V4 Promax87.3%60.82026-09-01
33Claude Opus 4.187.2%51.52026-09-01
34Qwen3.5 Plusthinking87.2%57.32026-09-01
35MiniMax-M2.187.0%53.82026-09-01
36GLM 5.186.9%55.92026-09-01
37GLM 5.3max86.8%63.52026-09-01
38GLM 5.286.7%52.02026-09-01
39GPT-5.6 Terraxhigh86.7%62.82026-09-01
40GPT-5high86.5%57.92026-09-01
41GPT-5.1high86.4%57.72026-09-01
42Grok 4.20thinking86.3%58.62026-09-01
43Gemini 3.1 Flash Lite86.2%53.72026-09-01
44GPT-5.2xhigh86.2%60.72026-09-01
45DeepSeek V4 Flashhigh86.2%57.42026-09-01
46Claude Opus 486.2%52.62026-09-01
47GLM 5.3 Flashmax86.1%54.22026-09-01
48GPT-5.6 Lunamax86.0%59.92026-09-01
49GLM-5thinking86.0%58.92026-09-01
50Kimi K2.5thinking85.9%57.62026-09-01
51Grok 4.3high85.8%57.02026-09-01
52Gemini 3.5 Flash Litehigh85.8%47.92026-09-01
53Nemotron 3 Ultra 550B A55B85.8%55.52026-09-01
54o3high85.6%53.92026-09-01
55Claude Opus 4.585.6%57.92026-09-01
56Inkling Small85.6%54.82026-09-01
57Grok 485.3%58.02026-09-01
58Qwen3 Maxmax85.0%52.82026-09-01
59DeepSeek V3.2thinking84.9%55.52026-09-01
60MiMo-V2.5-Pro84.6%54.92026-09-01

More knowledge benchmarks · All LLM benchmarks

Cite as: BenchLeader, “MMLU-Pro leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/vals_mmlu_pro.

MMLU-Pro: questions

What does MMLU-Pro measure?
MMLU-Pro is a harder successor to MMLU: about 12,000 questions across 14 subjects with ten answer options each and more reasoning-heavy items, removing the trivial and noisy questions of the original. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads MMLU-Pro?
Claude Fable 5.1 leads MMLU-Pro with 92.4% as of 24 Oct 2026, ahead of Claude Opus 5 at 91.6%.
How many models have MMLU-Pro results?
130 model configurations have a MMLU-Pro result on BenchLeader, all taken from Vals AI.
Who runs MMLU-Pro and how often is it updated?
MMLU-Pro is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does MMLU-Pro count toward the BenchLeader Index?
Yes. MMLU-Pro contributes to the knowledge category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.