Benchmark Almanac

KMMLU-Pro 2026

o1 78.1% leads since October 2026
Not in model cards
Version 2026

Models answer questions from the 2026 Korean national professional licensing exams. The dataset uses the KMMLU-Pro evaluation format.

The dataset also includes a nursing exam. The authors explain how to exclude nursing for comparisons with the original 14 licenses.

Made by
HAERAE-HUB
Released
October 2026
Kind
Knowledge in many languages
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
Korean
In and out
Text → Text
Interaction
Single turn
Items from
Exams and contests
Run by
Anyone
Based on
KMMLU-Pro
Occupations
Legal, Healthcare practitioners and technical
Strongest model tested
o3 #84 on the capability index
Last model added
July 2025
Scores refreshed
November 20th, 2026
26
Entries in its paper, 18 models we track

o1 leads with 78.1% among models released by July 2025 o1 leads with 78.1% among models released by July 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 o1 78.1%
  2. 2 Claude 3.7 Sonnet w/ thinking 77.7%
  3. 3 o3 73.6%
  4. 4 GPT-4.1 (2025-04-14) 73%
  5. 5 DeepSeek R1 (671B) 71.3%
  6. 6 o4-mini (2025-04-16) 69.7%
  7. 7 Grok 3 68.4%
  8. 8 Qwen3-235B-A22B w/ thinking 68.2%
  9. 9 Llama 4 Maverick 68.1%
  10. 10 EXAONE 4.0 32B w/ thinking 67.7%
  11. 11 Grok-3 mini 65.1%
  12. 12 QwQ-32B 63.9%
  13. 13 GPT-4.1 mini (2025-04-14) 62.2%
  14. 14 o3-mini 62%
  15. 15 Qwen3-32B w/ thinking 61.1%
  16. 16 DeepSeek V3 (671B) 60.8%
  17. 17 Qwen3-30B-A3B w/ thinking 60.5%
  18. 18 Qwen3-14B w/ thinking 60.2%
  19. 19 Llama 4 Scout 58.1%
  20. 20 C4AI Command A (111B) 57.5%
  21. 21 Llama 3.3 70B Instruct 53.2%
  22. 22 Gemma 3 27B 51%
  23. 23 Mistral Small 3.1 Instruct (24B) 49.5%
  24. 24 Gemma 3 12B 45.8%
  25. 25 Phi-4 (14B) 45.3%
  26. 26 Aya Expanse 32B 31.3%

Source: The KMMLU-Pro 2026 paper

#EntryScoreSettingDateSource
1 o1 78.1% Dec 2024 Official leaderboard
2 Claude 3.7 Sonnet 77.7% w/ thinking Feb 2025 Official leaderboard
3 o3 73.6% Apr 2025 Official leaderboard
4 GPT-4.1 (2025-04-14) 73% Official leaderboard
5 DeepSeek R1 (671B) 71.3% Official leaderboard
6 o4-mini (2025-04-16) 69.7% Official leaderboard
7 Grok 3 68.4% Apr 2025 Official leaderboard
8 Qwen3-235B-A22B 68.2% w/ thinking Apr 2025 Official leaderboard
9 Llama 4 Maverick 68.1% Apr 2025 Official leaderboard
10 EXAONE 4.0 32B 67.7% w/ thinking Jul 2025 Official leaderboard
11 Grok-3 mini 65.1% Apr 2025 Official leaderboard
12 QwQ-32B 63.9% Mar 2025 Official leaderboard
13 GPT-4.1 mini (2025-04-14) 62.2% Official leaderboard
14 o3-mini 62% Jan 2025 Official leaderboard
15 Qwen3-32B 61.1% w/ thinking Apr 2025 Official leaderboard
16 DeepSeek V3 (671B) 60.8% Official leaderboard
17 Qwen3-30B-A3B 60.5% w/ thinking Apr 2025 Official leaderboard
18 Qwen3-14B 60.2% w/ thinking Apr 2025 Official leaderboard
19 Llama 4 Scout 58.1% Apr 2025 Official leaderboard
20 C4AI Command A (111B) 57.5% Official leaderboard
21 Llama 3.3 70B Instruct 53.2% Dec 2024 Official leaderboard
22 Gemma 3 27B 51% Mar 2025 Official leaderboard
23 Mistral Small 3.1 Instruct (24B) 49.5% Official leaderboard
24 Gemma 3 12B 45.8% Mar 2025 Official leaderboard
25 Phi-4 (14B) 45.3% Official leaderboard
26 Aya Expanse 32B 31.3% Oct 2024 Official leaderboard
#ModelScoreSettingDateSource
1 o1 78.1% Dec 2024 Official leaderboard
2 Claude 3.7 Sonnet 77.7% w/ thinking Feb 2025 Official leaderboard
3 o3 73.6% Apr 2025 Official leaderboard
4 Grok 3 68.4% Apr 2025 Official leaderboard
5 Qwen3-235B-A22B 68.2% w/ thinking Apr 2025 Official leaderboard
6 Llama 4 Maverick 68.1% Apr 2025 Official leaderboard
7 EXAONE 4.0 32B 67.7% w/ thinking Jul 2025 Official leaderboard
8 Grok-3 mini 65.1% Apr 2025 Official leaderboard
9 QwQ-32B 63.9% Mar 2025 Official leaderboard
10 o3-mini 62% Jan 2025 Official leaderboard
11 Qwen3-32B 61.1% w/ thinking Apr 2025 Official leaderboard
12 Qwen3-30B-A3B 60.5% w/ thinking Apr 2025 Official leaderboard
13 Qwen3-14B 60.2% w/ thinking Apr 2025 Official leaderboard
14 Llama 4 Scout 58.1% Apr 2025 Official leaderboard
15 Llama 3.3 70B Instruct 53.2% Dec 2024 Official leaderboard
16 Gemma 3 27B 51% Mar 2025 Official leaderboard
17 Gemma 3 12B 45.8% Mar 2025 Official leaderboard
18 Aya Expanse 32B 31.3% Oct 2024 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3.7 Sonnet
w/ thinking
77.7%
Qwen3-235B-A22B
w/ thinking
68.2%
EXAONE 4.0 32B
w/ thinking
67.7%
Qwen3-32B
w/ thinking
61.1%
Qwen3-30B-A3B
w/ thinking
60.5%
Qwen3-14B
w/ thinking
60.2%

Source: The KMMLU-Pro 2026 paper

ModelEffortHarnessScoreCost per taskOutput tokens
o1 78.1%
Claude 3.7 Sonnet w/ thinking 77.7%
Claude 3.7 Sonnet 74.5%
o3 73.6%
Grok 3 68.4%
Qwen3-235B-A22B w/ thinking 68.2%
Llama 4 Maverick 68.1%
EXAONE 4.0 32B w/ thinking 67.7%
Grok-3 mini 65.1%
QwQ-32B 63.9%
Qwen3-235B-A22B 62.1%
o3-mini 62%
Qwen3-32B w/ thinking 61.1%
Qwen3-30B-A3B w/ thinking 60.5%
Qwen3-14B w/ thinking 60.2%
EXAONE 4.0 32B 60%
Qwen3-32B 58.9%
Llama 4 Scout 58.1%
Llama 3.3 70B Instruct 53.2%
Qwen3-14B 53%
Qwen3-30B-A3B 52.3%
Gemma 3 27B 51%
Gemma 3 12B 45.8%
Aya Expanse 32B 31.3%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3.7 Sonnet
w/ thinking
77.7%
Qwen3-235B-A22B
w/ thinking
68.2%
EXAONE 4.0 32B
w/ thinking
67.7%
Qwen3-32B
w/ thinking
61.1%
Qwen3-30B-A3B
w/ thinking
60.5%
Qwen3-14B
w/ thinking
60.2%

Source: The KMMLU-Pro 2026 paper

ModelEffortHarnessScoreCost per taskOutput tokens
o1 78.1%
Claude 3.7 Sonnet w/ thinking 77.7%
Claude 3.7 Sonnet 74.5%
o3 73.6%
Grok 3 68.4%
Qwen3-235B-A22B w/ thinking 68.2%
Llama 4 Maverick 68.1%
EXAONE 4.0 32B w/ thinking 67.7%
Grok-3 mini 65.1%
QwQ-32B 63.9%
Qwen3-235B-A22B 62.1%
o3-mini 62%
Qwen3-32B w/ thinking 61.1%
Qwen3-30B-A3B w/ thinking 60.5%
Qwen3-14B w/ thinking 60.2%
EXAONE 4.0 32B 60%
Qwen3-32B 58.9%
Llama 4 Scout 58.1%
Llama 3.3 70B Instruct 53.2%
Qwen3-14B 53%
Qwen3-30B-A3B 52.3%
Gemma 3 27B 51%
Gemma 3 12B 45.8%
Aya Expanse 32B 31.3%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Benchmark paper Not tested since Oct 2026
o1 78.1%

Source: The KMMLU-Pro 2026 paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o1 Llama 3.3 70B Instruct Aya Expanse 32B 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o1 Llama 3.3 70B Instruct

Source: Official KMMLU-Pro 2026 leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

Versions

More in Multilingual

All 545 in Multilingual
  • MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
  • WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET. K-EXAONE 90.5%
  • MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language. Qwen3.7-Max 87%
  • CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories. Solar Open 2 90.7%
  • Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
  • MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 16 months ago, in July 2025, for EXAONE 4.0 32B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (26 entries).
  • Pass Models scored 18 models we track, plus 8 other entries on its leaderboard.
  • Warning Newest score The newest came 16 months ago, in July 2025, for EXAONE 4.0 32B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o3 (#84).
  • Pass Headroom The best is 78.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 24 rows for models we track and 8 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check