Benchmark Almanac

PerMMLU

Gemini 2.5 Pro Jun 2025 84% leads since June 2025
Not in model cards

Models answer Persian multiple-choice questions from school curricula, university entrance exams, and Iranian cultural sources.

MIZAN ranks 45 model configurations. The test includes 80 cinema questions, 40 poetry questions, and 100 questions about proverbs.

Made by
MCINext
Released
May 2025
Kind
Knowledge in many languages
Level
School
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
Persian
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
o3 #84 on the capability index
Last model added
September 2025
Scores refreshed
November 20th, 2026
41
Entries on its own leaderboard, 35 models we track

Gemini 2.5 Pro leads with 84% among models released by September 2025 Gemini 2.5 Pro leads with 84% among models released by September 2025

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 September 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 Gemini 2.5 Pro Jun 2025, Thinking 84%
  2. 2 o3 Thinking 82.2%
  3. 3 GPT-5 mini Thinking 76.2%
  4. 4 Gemini 2.5 Flash Sep 2025, Thinking 72.8%
  5. 5 Claude 3.7 Sonnet Non-thinking 72.2%
  6. 6 GPT-4.1 Non-thinking 70.4%
  7. 7 Gemini 2.0 Flash Feb 2025, Non-thinking 69.6%
  8. 8 GPT-4o Mar 2025, Non-thinking 68.8%
  9. 9 GPT-5 nano Thinking 68.6%
  10. 10 deepseek-reasoner Thinking 68.1%
  11. 11 gpt-oss-120b Thinking 67.9%
  12. 12 Gemini 2.0 Flash-Lite Non-thinking 64.7%
  13. 13 GPT-4.1 mini Non-thinking 61.3%
  14. 14 Cohere Command A Non-thinking 59.8%
  15. 15 gpt-oss-20b Thinking 59.6%
  16. 16 deepseek-chat Non-thinking 59.1%
  17. 17 Llama 3.3 70B Instruct Non-thinking 57.1%
  18. 18 GPT-4o mini Non-thinking 57%
  19. 19 Llama 4 Scout Non-thinking 56.4%
  20. 20 Qwen3-32B Non-thinking 56.4%
  21. 21 Gemma 3 27B Non-thinking 56.3%
  22. 22 Claude 3.5 Haiku Non-thinking 54.4%
  23. 23 Qwen3-14B Non-thinking 51.4%
  24. 24 Gemma 3 12B Non-thinking 51%
  25. 25 Qwen3-30B-A3B Non-thinking 51%
  26. 26 deepseek-r1-0528-qwen3-8b Thinking 49.8%
  27. 27 Command R+ Non-thinking 48%
  28. 28 GPT-4.1 nano Non-thinking 47.8%
  29. 29 Mistral Small 3.1 Non-thinking 47.6%
  30. 30 Gemma 3n E4B Non-thinking 46.8%
  31. 31 Qwen3-8B Non-thinking 44.7%
  32. 32 Aya Expanse 32B Non-thinking 41.4%
  33. 33 Command R (03-2024) Non-thinking 40.5%
  34. 34 Qwen3-4B Non-thinking 40.3%
  35. 35 Gemma 3 4B Non-thinking 38.3%
  36. 36 yandexgpt-lite Non-thinking 35.3%
  37. 37 Gemma 3 1B Non-thinking 28.3%
  38. 38 Llama-3.2-3B-Instruct Non-thinking 27.8%
  39. 39 Mistral-7B-Instruct-v0.3 Non-thinking 25.5%
  40. 40 Llama-3.2-1B-Instruct Non-thinking 19.9%
  41. 41 Gemma 3 270M Non-thinking 19.5%

Source: Official PerMMLU leaderboard

Source: Official PerMMLU leaderboard

#EntryScoreSettingDateSource
1 Gemini 2.5 Pro Jun 2025 84% Thinking Jun 2025 Official leaderboard
2 o3 82.2% Thinking Apr 2025 Official leaderboard
3 GPT-5 mini 76.2% Thinking Aug 2025 Official leaderboard
4 Gemini 2.5 Flash Sep 2025 72.8% Thinking Sep 2025 Official leaderboard
5 Claude 3.7 Sonnet 72.2% Non-thinking Feb 2025 Official leaderboard
6 GPT-4.1 70.4% Non-thinking Apr 2025 Official leaderboard
7 Gemini 2.0 Flash Feb 2025 69.6% Non-thinking Feb 2025 Official leaderboard
8 GPT-4o Mar 2025 68.8% Non-thinking Mar 2025 Official leaderboard
9 GPT-5 nano 68.6% Thinking Aug 2025 Official leaderboard
10 deepseek-reasoner open weights, 671B 68.1% Thinking Official leaderboard
11 gpt-oss-120b 67.9% Thinking Aug 2025 Official leaderboard
12 Gemini 2.0 Flash-Lite 64.7% Non-thinking Feb 2025 Official leaderboard
13 GPT-4.1 mini 61.3% Non-thinking Apr 2025 Official leaderboard
14 Cohere Command A 59.8% Non-thinking Mar 2025 Official leaderboard
15 gpt-oss-20b 59.6% Thinking Aug 2025 Official leaderboard
16 deepseek-chat open weights, 671B 59.1% Non-thinking Official leaderboard
17 Llama 3.3 70B Instruct 57.1% Non-thinking Dec 2024 Official leaderboard
18 GPT-4o mini 57% Non-thinking Jul 2024 Official leaderboard
19 Llama 4 Scout 56.4% Non-thinking Apr 2025 Official leaderboard
20 Qwen3-32B 56.4% Non-thinking Apr 2025 Official leaderboard
21 Gemma 3 27B 56.3% Non-thinking Mar 2025 Official leaderboard
22 Claude 3.5 Haiku 54.4% Non-thinking Oct 2024 Official leaderboard
23 Qwen3-14B 51.4% Non-thinking Apr 2025 Official leaderboard
24 Gemma 3 12B 51% Non-thinking Mar 2025 Official leaderboard
25 Qwen3-30B-A3B 51% Non-thinking Apr 2025 Official leaderboard
26 deepseek-r1-0528-qwen3-8b 49.8% Thinking May 2025 Official leaderboard
27 Command R+ 48% Non-thinking Aug 2024 Official leaderboard
28 GPT-4.1 nano 47.8% Non-thinking Apr 2025 Official leaderboard
29 Mistral Small 3.1 47.6% Non-thinking Mar 2025 Official leaderboard
30 Gemma 3n E4B 46.8% Non-thinking Jun 2025 Official leaderboard
31 Qwen3-8B 44.7% Non-thinking Apr 2025 Official leaderboard
32 Aya Expanse 32B 41.4% Non-thinking Oct 2024 Official leaderboard
33 Command R (03-2024) 40.5% Non-thinking Mar 2024 Official leaderboard
34 Qwen3-4B 40.3% Non-thinking Apr 2025 Official leaderboard
35 Gemma 3 4B 38.3% Non-thinking Mar 2025 Official leaderboard
36 yandexgpt-lite open weights, 8B 35.3% Non-thinking Official leaderboard
37 Gemma 3 1B 28.3% Non-thinking Mar 2025 Official leaderboard
38 Llama-3.2-3B-Instruct open weights, 3.2B 27.8% Non-thinking Official leaderboard
39 Mistral-7B-Instruct-v0.3 open weights, 7.3B 25.5% Non-thinking Official leaderboard
40 Llama-3.2-1B-Instruct open weights, 1.2B 19.9% Non-thinking Official leaderboard
41 Gemma 3 270M 19.5% Non-thinking Aug 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 2.5 Pro Jun 2025 84% Thinking Jun 2025 Official leaderboard
2 o3 82.2% Thinking Apr 2025 Official leaderboard
3 GPT-5 mini 76.2% Thinking Aug 2025 Official leaderboard
4 Gemini 2.5 Flash Sep 2025 72.8% Thinking Sep 2025 Official leaderboard
5 Claude 3.7 Sonnet 72.2% Non-thinking Feb 2025 Official leaderboard
6 GPT-4.1 70.4% Non-thinking Apr 2025 Official leaderboard
7 Gemini 2.0 Flash Feb 2025 69.6% Non-thinking Feb 2025 Official leaderboard
8 GPT-4o Mar 2025 68.8% Non-thinking Mar 2025 Official leaderboard
9 GPT-5 nano 68.6% Thinking Aug 2025 Official leaderboard
10 gpt-oss-120b 67.9% Thinking Aug 2025 Official leaderboard
11 Gemini 2.0 Flash-Lite 64.7% Non-thinking Feb 2025 Official leaderboard
12 GPT-4.1 mini 61.3% Non-thinking Apr 2025 Official leaderboard
13 Cohere Command A 59.8% Non-thinking Mar 2025 Official leaderboard
14 gpt-oss-20b 59.6% Thinking Aug 2025 Official leaderboard
15 Llama 3.3 70B Instruct 57.1% Non-thinking Dec 2024 Official leaderboard
16 GPT-4o mini 57% Non-thinking Jul 2024 Official leaderboard
17 Llama 4 Scout 56.4% Non-thinking Apr 2025 Official leaderboard
18 Qwen3-32B 56.4% Non-thinking Apr 2025 Official leaderboard
19 Gemma 3 27B 56.3% Non-thinking Mar 2025 Official leaderboard
20 Claude 3.5 Haiku 54.4% Non-thinking Oct 2024 Official leaderboard
21 Qwen3-14B 51.4% Non-thinking Apr 2025 Official leaderboard
22 Gemma 3 12B 51% Non-thinking Mar 2025 Official leaderboard
23 Qwen3-30B-A3B 51% Non-thinking Apr 2025 Official leaderboard
24 deepseek-r1-0528-qwen3-8b 49.8% Thinking May 2025 Official leaderboard
25 Command R+ 48% Non-thinking Aug 2024 Official leaderboard
26 GPT-4.1 nano 47.8% Non-thinking Apr 2025 Official leaderboard
27 Mistral Small 3.1 47.6% Non-thinking Mar 2025 Official leaderboard
28 Gemma 3n E4B 46.8% Non-thinking Jun 2025 Official leaderboard
29 Qwen3-8B 44.7% Non-thinking Apr 2025 Official leaderboard
30 Aya Expanse 32B 41.4% Non-thinking Oct 2024 Official leaderboard
31 Command R (03-2024) 40.5% Non-thinking Mar 2024 Official leaderboard
32 Qwen3-4B 40.3% Non-thinking Apr 2025 Official leaderboard
33 Gemma 3 4B 38.3% Non-thinking Mar 2025 Official leaderboard
34 Gemma 3 1B 28.3% Non-thinking Mar 2025 Official leaderboard
35 Gemma 3 270M 19.5% Non-thinking Aug 2025 Official leaderboard

Rankings by part

We read 18 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as cinema acc, emergency number acc and foods acc. Gemini 2.5 Pro leads 10 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

cinema acc

35 models we track have a score here.

  1. 1 Gemini 2.5 Pro Jun 2025 93.8%
  2. 2 Gemini 2.5 Flash Sep 2025 ↑2 88.8%
  3. 3 o3 ↓1 81.3%
  4. 4 GPT-4.1 ↑2 76.3%
  5. 5 GPT-4o Mar 2025 ↑3 75%
  6. 6 Gemini 2.0 Flash Feb 2025 ↑1 72.5%
  7. 7 Gemini 2.0 Flash-Lite ↑4 71.3%
  8. 8 Claude 3.7 Sonnet ↓3 68.8%
  9. 9 Command R+ ↑16 67.5%
  10. 10 GPT-4o mini ↑6 66.3%

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
GPT-5 mini
Thinking minimal rea…
76.2%
GPT-5 nano
Thinking minimal rea…
68.6%
gpt-oss-120b
Thinking low reasoni…
67.9%
gpt-oss-20b
Thinking low reasoni…
59.6%

Source: Official PerMMLU leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 2.5 Pro Jun 2025 Thinking 84%
o3 Thinking 82.2%
GPT-5 mini Thinking 76.2%
Gemini 2.5 Flash Sep 2025 Thinking 72.8%
Claude 3.7 Sonnet Non-thinking 72.2%
GPT-4.1 Non-thinking 70.4%
Gemini 2.0 Flash Feb 2025 Non-thinking 69.6%
GPT-4o Mar 2025 Non-thinking 68.8%
GPT-5 nano Thinking 68.6%
gpt-oss-120b Thinking 67.9%
Gemini 2.0 Flash-Lite Non-thinking 64.7%
gpt-oss-120b low reasoning 62.2%
GPT-4.1 mini Non-thinking 61.3%
Cohere Command A Non-thinking 59.8%
gpt-oss-20b Thinking 59.6%
GPT-5 mini minimal reasoning 59.5%
Llama 3.3 70B Instruct Non-thinking 57.1%
GPT-4o mini Non-thinking 57%
Llama 4 Scout Non-thinking 56.4%
Qwen3-32B Non-thinking 56.4%
Gemma 3 27B Non-thinking 56.3%
Claude 3.5 Haiku Non-thinking 54.4%
gpt-oss-20b low reasoning 54.4%
Qwen3-14B Non-thinking 51.4%
Gemma 3 12B Non-thinking 51%
Qwen3-30B-A3B Non-thinking 51%
deepseek-r1-0528-qwen3-8b Thinking 49.8%
Command R+ Non-thinking 48%
GPT-4.1 nano Non-thinking 47.8%
Mistral Small 3.1 Non-thinking 47.6%
Gemma 3n E4B Non-thinking 46.8%
GPT-5 nano minimal reasoning 45.6%
Qwen3-8B Non-thinking 44.7%
Aya Expanse 32B Non-thinking 41.4%
Command R (03-2024) Non-thinking 40.5%
Qwen3-4B Non-thinking 40.3%
Gemma 3 4B Non-thinking 38.3%
Gemma 3 1B Non-thinking 28.3%
Gemma 3 270M Non-thinking 19.5%

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
GPT-5 mini
Thinking minimal rea…
76.2%
GPT-5 nano
Thinking minimal rea…
68.6%
gpt-oss-120b
Thinking low reasoni…
67.9%
gpt-oss-20b
Thinking low reasoni…
59.6%

Source: Official PerMMLU leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 2.5 Pro Jun 2025 Thinking 84%
o3 Thinking 82.2%
GPT-5 mini Thinking 76.2%
Gemini 2.5 Flash Sep 2025 Thinking 72.8%
Claude 3.7 Sonnet Non-thinking 72.2%
GPT-4.1 Non-thinking 70.4%
Gemini 2.0 Flash Feb 2025 Non-thinking 69.6%
GPT-4o Mar 2025 Non-thinking 68.8%
GPT-5 nano Thinking 68.6%
gpt-oss-120b Thinking 67.9%
Gemini 2.0 Flash-Lite Non-thinking 64.7%
gpt-oss-120b low reasoning 62.2%
GPT-4.1 mini Non-thinking 61.3%
Cohere Command A Non-thinking 59.8%
gpt-oss-20b Thinking 59.6%
GPT-5 mini minimal reasoning 59.5%
Llama 3.3 70B Instruct Non-thinking 57.1%
GPT-4o mini Non-thinking 57%
Llama 4 Scout Non-thinking 56.4%
Qwen3-32B Non-thinking 56.4%
Gemma 3 27B Non-thinking 56.3%
Claude 3.5 Haiku Non-thinking 54.4%
gpt-oss-20b low reasoning 54.4%
Qwen3-14B Non-thinking 51.4%
Gemma 3 12B Non-thinking 51%
Qwen3-30B-A3B Non-thinking 51%
deepseek-r1-0528-qwen3-8b Thinking 49.8%
Command R+ Non-thinking 48%
GPT-4.1 nano Non-thinking 47.8%
Mistral Small 3.1 Non-thinking 47.6%
Gemma 3n E4B Non-thinking 46.8%
GPT-5 nano minimal reasoning 45.6%
Qwen3-8B Non-thinking 44.7%
Aya Expanse 32B Non-thinking 41.4%
Command R (03-2024) Non-thinking 40.5%
Qwen3-4B Non-thinking 40.3%
Gemma 3 4B Non-thinking 38.3%
Gemma 3 1B Non-thinking 28.3%
Gemma 3 270M Non-thinking 19.5%

2 models have held number one on the official leaderboard since its release in May 2025

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

OpenAI Google DeepMind
Own leaderboard Not tested since Sep 2025

Source: Official PerMMLU leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Gemini 2.5 Pro Claude 3.7 Sonnet Llama 3.3 70B Instruct GPT-4o mini Command R (03-2024) 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Gemini 2.5 Pro Claude 3.7 Sonnet

Source: Official PerMMLU 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.

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 14 months ago, in September 2025, for Gemini 2.5 Flash. 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 official leaderboard (41 entries).
  • Pass Models scored 35 models we track, plus 6 other entries on its leaderboard.
  • Warning Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o3 (#84).
  • Pass Headroom The best is 84%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on mcinext-mizan-llm-leaderboard.hf.space: 39 rows for models we track and 6 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