Benchmark Almanac

Persian MT-Bench

o3 9 leads since May 2025
Not in model cards

Models follow Persian conversations across two to four turns. One asks for film directors, then actors, then the oldest and newest films.

MIZAN ranks 45 model configurations on 100 dialogues. Ten dialogues cover Iranian cultural knowledge and ten test retrieval during conversation.

Made by
MCINext
Released
May 2025
Kind
Chat and instructions in many languages
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
Persian
In and out
Text, Code → Text, Code
Interaction
Multi-turn
Items from
Another benchmark
Refresh
One fixed set
Run by
Anyone
Based on
MT-Bench
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

o3 leads with 9 among models released by September 2025 o3 leads with 9 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 o3 Thinking 9
  2. 2 Gemini 2.5 Pro Jun 2025, Thinking 9
  3. 3 Gemini 2.5 Flash Sep 2025, Thinking 9
  4. 4 GPT-4.1 Non-thinking 9
  5. 5 Claude 3.7 Sonnet Non-thinking 9
  6. 6 GPT-5 mini Thinking 9
  7. 7 deepseek-reasoner Thinking 9
  8. 8 deepseek-chat Non-thinking 9
  9. 9 gpt-oss-120b Thinking 8
  10. 10 GPT-4.1 mini Non-thinking 8
  11. 11 Gemini 2.0 Flash Feb 2025, Non-thinking 8
  12. 12 GPT-4o Mar 2025, Non-thinking 8
  13. 13 GPT-5 nano Thinking 8
  14. 14 Gemini 2.0 Flash-Lite Non-thinking 8
  15. 15 Cohere Command A Non-thinking 8
  16. 16 gpt-oss-20b Thinking 8
  17. 17 Gemma 3 27B Non-thinking 8
  18. 18 GPT-4o mini Non-thinking 8
  19. 19 Qwen3-32B Non-thinking 8
  20. 20 Claude 3.5 Haiku Non-thinking 8
  21. 21 Gemma 3 12B Non-thinking 8
  22. 22 Qwen3-30B-A3B Non-thinking 7
  23. 23 Llama 4 Scout Non-thinking 7
  24. 24 GPT-4.1 nano Non-thinking 7
  25. 25 Mistral Small 3.1 Non-thinking 7
  26. 26 Qwen3-14B Non-thinking 7
  27. 27 Llama 3.3 70B Instruct Non-thinking 7
  28. 28 Aya Expanse 32B Non-thinking 7
  29. 29 Command R+ Non-thinking 7
  30. 30 Gemma 3n E4B Non-thinking 7
  31. 31 deepseek-r1-0528-qwen3-8b Thinking 7
  32. 32 Gemma 3 4B Non-thinking 7
  33. 33 Qwen3-8B Non-thinking 7
  34. 34 Command R (03-2024) Non-thinking 6
  35. 35 Qwen3-4B Non-thinking 6
  36. 36 yandexgpt-lite Non-thinking 5
  37. 37 Gemma 3 1B Non-thinking 4
  38. 38 Llama-3.2-3B-Instruct Non-thinking 4
  39. 39 Mistral-7B-Instruct-v0.3 Non-thinking 3
  40. 40 Llama-3.2-1B-Instruct Non-thinking 3
  41. 41 Gemma 3 270M Non-thinking 2

Source: Official Persian MT-Bench leaderboard

Source: Official Persian MT-Bench leaderboard

#EntryScoreSettingDateSource
1 o3 9 Thinking Apr 2025 Official leaderboard
2 Gemini 2.5 Pro Jun 2025 9 Thinking Jun 2025 Official leaderboard
3 Gemini 2.5 Flash Sep 2025 9 Thinking Sep 2025 Official leaderboard
4 GPT-4.1 9 Non-thinking Apr 2025 Official leaderboard
5 Claude 3.7 Sonnet 9 Non-thinking Feb 2025 Official leaderboard
6 GPT-5 mini 9 Thinking Aug 2025 Official leaderboard
7 deepseek-reasoner open weights, 671B 9 Thinking Official leaderboard
8 deepseek-chat open weights, 671B 9 Non-thinking Official leaderboard
9 gpt-oss-120b 8 Thinking Aug 2025 Official leaderboard
10 GPT-4.1 mini 8 Non-thinking Apr 2025 Official leaderboard
11 Gemini 2.0 Flash Feb 2025 8 Non-thinking Feb 2025 Official leaderboard
12 GPT-4o Mar 2025 8 Non-thinking Mar 2025 Official leaderboard
13 GPT-5 nano 8 Thinking Aug 2025 Official leaderboard
14 Gemini 2.0 Flash-Lite 8 Non-thinking Feb 2025 Official leaderboard
15 Cohere Command A 8 Non-thinking Mar 2025 Official leaderboard
16 gpt-oss-20b 8 Thinking Aug 2025 Official leaderboard
17 Gemma 3 27B 8 Non-thinking Mar 2025 Official leaderboard
18 GPT-4o mini 8 Non-thinking Jul 2024 Official leaderboard
19 Qwen3-32B 8 Non-thinking Apr 2025 Official leaderboard
20 Claude 3.5 Haiku 8 Non-thinking Oct 2024 Official leaderboard
21 Gemma 3 12B 8 Non-thinking Mar 2025 Official leaderboard
22 Qwen3-30B-A3B 7 Non-thinking Apr 2025 Official leaderboard
23 Llama 4 Scout 7 Non-thinking Apr 2025 Official leaderboard
24 GPT-4.1 nano 7 Non-thinking Apr 2025 Official leaderboard
25 Mistral Small 3.1 7 Non-thinking Mar 2025 Official leaderboard
26 Qwen3-14B 7 Non-thinking Apr 2025 Official leaderboard
27 Llama 3.3 70B Instruct 7 Non-thinking Dec 2024 Official leaderboard
28 Aya Expanse 32B 7 Non-thinking Oct 2024 Official leaderboard
29 Command R+ 7 Non-thinking Aug 2024 Official leaderboard
30 Gemma 3n E4B 7 Non-thinking Jun 2025 Official leaderboard
31 deepseek-r1-0528-qwen3-8b 7 Thinking May 2025 Official leaderboard
32 Gemma 3 4B 7 Non-thinking Mar 2025 Official leaderboard
33 Qwen3-8B 7 Non-thinking Apr 2025 Official leaderboard
34 Command R (03-2024) 6 Non-thinking Mar 2024 Official leaderboard
35 Qwen3-4B 6 Non-thinking Apr 2025 Official leaderboard
36 yandexgpt-lite open weights, 8B 5 Non-thinking Official leaderboard
37 Gemma 3 1B 4 Non-thinking Mar 2025 Official leaderboard
38 Llama-3.2-3B-Instruct open weights, 3.2B 4 Non-thinking Official leaderboard
39 Mistral-7B-Instruct-v0.3 open weights, 7.3B 3 Non-thinking Official leaderboard
40 Llama-3.2-1B-Instruct open weights, 1.2B 3 Non-thinking Official leaderboard
41 Gemma 3 270M 2 Non-thinking Aug 2025 Official leaderboard
#ModelScoreSettingDateSource
1 o3 9 Thinking Apr 2025 Official leaderboard
2 Gemini 2.5 Pro Jun 2025 9 Thinking Jun 2025 Official leaderboard
3 Gemini 2.5 Flash Sep 2025 9 Thinking Sep 2025 Official leaderboard
4 GPT-4.1 9 Non-thinking Apr 2025 Official leaderboard
5 Claude 3.7 Sonnet 9 Non-thinking Feb 2025 Official leaderboard
6 GPT-5 mini 9 Thinking Aug 2025 Official leaderboard
7 gpt-oss-120b 8 Thinking Aug 2025 Official leaderboard
8 GPT-4.1 mini 8 Non-thinking Apr 2025 Official leaderboard
9 Gemini 2.0 Flash Feb 2025 8 Non-thinking Feb 2025 Official leaderboard
10 GPT-4o Mar 2025 8 Non-thinking Mar 2025 Official leaderboard
11 GPT-5 nano 8 Thinking Aug 2025 Official leaderboard
12 Gemini 2.0 Flash-Lite 8 Non-thinking Feb 2025 Official leaderboard
13 Cohere Command A 8 Non-thinking Mar 2025 Official leaderboard
14 gpt-oss-20b 8 Thinking Aug 2025 Official leaderboard
15 Gemma 3 27B 8 Non-thinking Mar 2025 Official leaderboard
16 GPT-4o mini 8 Non-thinking Jul 2024 Official leaderboard
17 Qwen3-32B 8 Non-thinking Apr 2025 Official leaderboard
18 Claude 3.5 Haiku 8 Non-thinking Oct 2024 Official leaderboard
19 Gemma 3 12B 8 Non-thinking Mar 2025 Official leaderboard
20 Qwen3-30B-A3B 7 Non-thinking Apr 2025 Official leaderboard
21 Llama 4 Scout 7 Non-thinking Apr 2025 Official leaderboard
22 GPT-4.1 nano 7 Non-thinking Apr 2025 Official leaderboard
23 Mistral Small 3.1 7 Non-thinking Mar 2025 Official leaderboard
24 Qwen3-14B 7 Non-thinking Apr 2025 Official leaderboard
25 Llama 3.3 70B Instruct 7 Non-thinking Dec 2024 Official leaderboard
26 Aya Expanse 32B 7 Non-thinking Oct 2024 Official leaderboard
27 Command R+ 7 Non-thinking Aug 2024 Official leaderboard
28 Gemma 3n E4B 7 Non-thinking Jun 2025 Official leaderboard
29 deepseek-r1-0528-qwen3-8b 7 Thinking May 2025 Official leaderboard
30 Gemma 3 4B 7 Non-thinking Mar 2025 Official leaderboard
31 Qwen3-8B 7 Non-thinking Apr 2025 Official leaderboard
32 Command R (03-2024) 6 Non-thinking Mar 2024 Official leaderboard
33 Qwen3-4B 6 Non-thinking Apr 2025 Official leaderboard
34 Gemma 3 1B 4 Non-thinking Mar 2025 Official leaderboard
35 Gemma 3 270M 2 Non-thinking Aug 2025 Official leaderboard

Rankings by part

We read 21 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as score w mean, writing score w mean and writing score mean. o3 leads 9 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

score w mean

35 models we track have a score here.

  1. 1 o3 9
  2. 2 Gemini 2.5 Pro Jun 2025 9
  3. 3 Gemini 2.5 Flash Sep 2025 9
  4. 4 Claude 3.7 Sonnet ↑1 9
  5. 5 GPT-4.1 ↓1 9
  6. 6 GPT-5 mini 9
  7. 7 Gemini 2.0 Flash Feb 2025 ↑2 8
  8. 8 gpt-oss-120b ↓1 8
  9. 9 GPT-4o Mar 2025 ↑1 8
  10. 10 GPT-4.1 mini ↓2 8

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…
9
gpt-oss-120b
Thinking low reasoni…
8
GPT-5 nano
Thinking minimal rea…
8
gpt-oss-20b
Thinking low reasoni…
8

Source: Official Persian MT-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
o3 Thinking 9
Gemini 2.5 Pro Jun 2025 Thinking 9
Gemini 2.5 Flash Sep 2025 Thinking 9
GPT-4.1 Non-thinking 9
Claude 3.7 Sonnet Non-thinking 9
GPT-5 mini Thinking 9
gpt-oss-120b Thinking 8
GPT-4.1 mini Non-thinking 8
Gemini 2.0 Flash Feb 2025 Non-thinking 8
GPT-4o Mar 2025 Non-thinking 8
GPT-5 mini minimal reasoning 8
GPT-5 nano Thinking 8
Gemini 2.0 Flash-Lite Non-thinking 8
Cohere Command A Non-thinking 8
gpt-oss-20b Thinking 8
Gemma 3 27B Non-thinking 8
gpt-oss-120b low reasoning 8
GPT-4o mini Non-thinking 8
Qwen3-32B Non-thinking 8
Claude 3.5 Haiku Non-thinking 8
Gemma 3 12B Non-thinking 8
GPT-5 nano minimal reasoning 7
Qwen3-30B-A3B Non-thinking 7
Llama 4 Scout Non-thinking 7
gpt-oss-20b low reasoning 7
GPT-4.1 nano Non-thinking 7
Mistral Small 3.1 Non-thinking 7
Qwen3-14B Non-thinking 7
Llama 3.3 70B Instruct Non-thinking 7
Aya Expanse 32B Non-thinking 7
Command R+ Non-thinking 7
Gemma 3n E4B Non-thinking 7
deepseek-r1-0528-qwen3-8b Thinking 7
Gemma 3 4B Non-thinking 7
Qwen3-8B Non-thinking 7
Command R (03-2024) Non-thinking 6
Qwen3-4B Non-thinking 6
Gemma 3 1B Non-thinking 4
Gemma 3 270M Non-thinking 2

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…
9
gpt-oss-120b
Thinking low reasoni…
8
GPT-5 nano
Thinking minimal rea…
8
gpt-oss-20b
Thinking low reasoni…
8

Source: Official Persian MT-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
o3 Thinking 9
Gemini 2.5 Pro Jun 2025 Thinking 9
Gemini 2.5 Flash Sep 2025 Thinking 9
GPT-4.1 Non-thinking 9
Claude 3.7 Sonnet Non-thinking 9
GPT-5 mini Thinking 9
gpt-oss-120b Thinking 8
GPT-4.1 mini Non-thinking 8
Gemini 2.0 Flash Feb 2025 Non-thinking 8
GPT-4o Mar 2025 Non-thinking 8
GPT-5 mini minimal reasoning 8
GPT-5 nano Thinking 8
Gemini 2.0 Flash-Lite Non-thinking 8
Cohere Command A Non-thinking 8
gpt-oss-20b Thinking 8
Gemma 3 27B Non-thinking 8
gpt-oss-120b low reasoning 8
GPT-4o mini Non-thinking 8
Qwen3-32B Non-thinking 8
Claude 3.5 Haiku Non-thinking 8
Gemma 3 12B Non-thinking 8
GPT-5 nano minimal reasoning 7
Qwen3-30B-A3B Non-thinking 7
Llama 4 Scout Non-thinking 7
gpt-oss-20b low reasoning 7
GPT-4.1 nano Non-thinking 7
Mistral Small 3.1 Non-thinking 7
Qwen3-14B Non-thinking 7
Llama 3.3 70B Instruct Non-thinking 7
Aya Expanse 32B Non-thinking 7
Command R+ Non-thinking 7
Gemma 3n E4B Non-thinking 7
deepseek-r1-0528-qwen3-8b Thinking 7
Gemma 3 4B Non-thinking 7
Qwen3-8B Non-thinking 7
Command R (03-2024) Non-thinking 6
Qwen3-4B Non-thinking 6
Gemma 3 1B Non-thinking 4
Gemma 3 270M Non-thinking 2

Number one over time

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

OpenAI
Own leaderboard Not tested since Sep 2025
o3 9

Source: Official Persian MT-Bench 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
0 2 4 6 8 10 2024 2025 0 2 4 6 8 10 2024 2025

Source: Official Persian MT-Bench 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

7 of 9 checks pass, 2 warnings. Look first at the Confidence check. Likely. Some facts rest on secondary sources. Every benchmark

Catalog record 1 warning 1 pass
  • Pass Task access All tasks are public.
  • Warning Confidence Likely. Some facts rest on secondary sources.
Scores 1 warning 2 pass 2 notes
  • 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).
  • Note Headroom Scored as a rating, with no fixed ceiling.
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