Benchmark Almanac

Persian NLG

Llama 3.3 70B Instruct 20.1 leads since May 2025
Not in model cards

Models summarize Persian texts, translate between Persian and two other languages, and write questions from passages.

MIZAN ranks 45 model configurations. Its English translation tests include literary passages, subtitle dialogue, and religious texts in both directions.

Made by
MCINext
Released
May 2025
Kind
Translation
Level
University
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
Persian, English, Arabic
In and out
Text → Text
Interaction
Single turn
Items from
Another benchmark
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, 37 models we track

Llama 3.3 70B Instruct leads with 20.1 among models released by September 2025 Llama 3.3 70B Instruct leads with 20.1 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 Llama 3.3 70B Instruct Non-thinking 20.1
  2. 2 GPT-4.1 Non-thinking 19.5
  3. 3 GPT-4.1 mini Non-thinking 19.0
  4. 4 GPT-4o Mar 2025, Non-thinking 19.0
  5. 5 Command R+ Non-thinking 18.8
  6. 6 GPT-5 mini Thinking 18.2
  7. 7 GPT-4o mini Non-thinking 18.1
  8. 8 Gemini 2.0 Flash Feb 2025, Non-thinking 17.8
  9. 9 Claude 3.7 Sonnet Non-thinking 17.8
  10. 10 o3 Thinking 17.6
  11. 11 Qwen3-32B Non-thinking 16.8
  12. 12 GPT-4.1 nano Non-thinking 16.7
  13. 13 Gemini 2.0 Flash-Lite Non-thinking 16.6
  14. 14 GPT-5 nano Thinking 16.4
  15. 15 Command R (03-2024) Non-thinking 16.4
  16. 16 Qwen3-30B-A3B Non-thinking 16.4
  17. 17 Qwen3-14B Non-thinking 16.1
  18. 18 Llama 4 Scout Non-thinking 15.7
  19. 19 Qwen3-8B Non-thinking 15.6
  20. 20 gpt-oss-120b Thinking 15.4
  21. 21 Qwen3-4B Non-thinking 13.9
  22. 22 Gemini 2.5 Flash Sep 2025, Thinking 13.7
  23. 23 gpt-oss-20b Thinking 13.3
  24. 24 Mistral Small 3.1 Non-thinking 13.2
  25. 25 Gemma 3 12B Non-thinking 12.0
  26. 26 Aya Expanse 32B Non-thinking 12.0
  27. 27 Gemini 2.5 Pro Jun 2025, Thinking 11.5
  28. 28 Cohere Command A Non-thinking 11.4
  29. 29 Llama 3.2 3B Non-thinking 11.3
  30. 30 deepseek-r1-0528-qwen3-8b Thinking 11.2
  31. 31 Claude 3.5 Haiku Non-thinking 10.9
  32. 32 Gemma 3 27B Non-thinking 10.7
  33. 33 yandexgpt-lite Non-thinking 10.3
  34. 34 Gemma 3 4B Non-thinking 9.50
  35. 35 Mistral-7B-Instruct-v0.3 Non-thinking 9.44
  36. 36 Gemma 3n E4B Non-thinking 9.40
  37. 37 deepseek-chat Non-thinking 9.34
  38. 38 deepseek-reasoner Thinking 8.81
  39. 39 Llama 3.2 1B Non-thinking 8.23
  40. 40 Gemma 3 1B Non-thinking 6.83
  41. 41 Gemma 3 270M Non-thinking 5.10

Source: Official Persian NLG leaderboard

Source: Official Persian NLG leaderboard

#EntryScoreSettingDateSource
1 Llama 3.3 70B Instruct 20.1 Non-thinking Dec 2024 Official leaderboard
2 GPT-4.1 19.5 Non-thinking Apr 2025 Official leaderboard
3 GPT-4.1 mini 19.0 Non-thinking Apr 2025 Official leaderboard
4 GPT-4o Mar 2025 19.0 Non-thinking Mar 2025 Official leaderboard
5 Command R+ 18.8 Non-thinking Aug 2024 Official leaderboard
6 GPT-5 mini 18.2 Thinking Aug 2025 Official leaderboard
7 GPT-4o mini 18.1 Non-thinking Jul 2024 Official leaderboard
8 Gemini 2.0 Flash Feb 2025 17.8 Non-thinking Feb 2025 Official leaderboard
9 Claude 3.7 Sonnet 17.8 Non-thinking Feb 2025 Official leaderboard
10 o3 17.6 Thinking Apr 2025 Official leaderboard
11 Qwen3-32B 16.8 Non-thinking Apr 2025 Official leaderboard
12 GPT-4.1 nano 16.7 Non-thinking Apr 2025 Official leaderboard
13 Gemini 2.0 Flash-Lite 16.6 Non-thinking Feb 2025 Official leaderboard
14 GPT-5 nano 16.4 Thinking Aug 2025 Official leaderboard
15 Command R (03-2024) 16.4 Non-thinking Mar 2024 Official leaderboard
16 Qwen3-30B-A3B 16.4 Non-thinking Apr 2025 Official leaderboard
17 Qwen3-14B 16.1 Non-thinking Apr 2025 Official leaderboard
18 Llama 4 Scout 15.7 Non-thinking Apr 2025 Official leaderboard
19 Qwen3-8B 15.6 Non-thinking Apr 2025 Official leaderboard
20 gpt-oss-120b 15.4 Thinking Aug 2025 Official leaderboard
21 Qwen3-4B 13.9 Non-thinking Apr 2025 Official leaderboard
22 Gemini 2.5 Flash Sep 2025 13.7 Thinking Sep 2025 Official leaderboard
23 gpt-oss-20b 13.3 Thinking Aug 2025 Official leaderboard
24 Mistral Small 3.1 13.2 Non-thinking Mar 2025 Official leaderboard
25 Gemma 3 12B 12.0 Non-thinking Mar 2025 Official leaderboard
26 Aya Expanse 32B 12.0 Non-thinking Oct 2024 Official leaderboard
27 Gemini 2.5 Pro Jun 2025 11.5 Thinking Jun 2025 Official leaderboard
28 Cohere Command A 11.4 Non-thinking Mar 2025 Official leaderboard
29 Llama 3.2 3B 11.3 Non-thinking Sep 2024 Official leaderboard
30 deepseek-r1-0528-qwen3-8b 11.2 Thinking May 2025 Official leaderboard
31 Claude 3.5 Haiku 10.9 Non-thinking Oct 2024 Official leaderboard
32 Gemma 3 27B 10.7 Non-thinking Mar 2025 Official leaderboard
33 yandexgpt-lite open weights, 8B 10.3 Non-thinking Official leaderboard
34 Gemma 3 4B 9.50 Non-thinking Mar 2025 Official leaderboard
35 Mistral-7B-Instruct-v0.3 open weights, 7.3B 9.44 Non-thinking Official leaderboard
36 Gemma 3n E4B 9.40 Non-thinking Jun 2025 Official leaderboard
37 deepseek-chat open weights, 671B 9.34 Non-thinking Official leaderboard
38 deepseek-reasoner open weights, 671B 8.81 Thinking Official leaderboard
39 Llama 3.2 1B 8.23 Non-thinking Sep 2024 Official leaderboard
40 Gemma 3 1B 6.83 Non-thinking Mar 2025 Official leaderboard
41 Gemma 3 270M 5.10 Non-thinking Aug 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Llama 3.3 70B Instruct 20.1 Non-thinking Dec 2024 Official leaderboard
2 GPT-4.1 19.5 Non-thinking Apr 2025 Official leaderboard
3 GPT-4.1 mini 19.0 Non-thinking Apr 2025 Official leaderboard
4 GPT-4o Mar 2025 19.0 Non-thinking Mar 2025 Official leaderboard
5 Command R+ 18.8 Non-thinking Aug 2024 Official leaderboard
6 GPT-5 mini 18.2 Thinking Aug 2025 Official leaderboard
7 GPT-4o mini 18.1 Non-thinking Jul 2024 Official leaderboard
8 Gemini 2.0 Flash Feb 2025 17.8 Non-thinking Feb 2025 Official leaderboard
9 Claude 3.7 Sonnet 17.8 Non-thinking Feb 2025 Official leaderboard
10 o3 17.6 Thinking Apr 2025 Official leaderboard
11 Qwen3-32B 16.8 Non-thinking Apr 2025 Official leaderboard
12 GPT-4.1 nano 16.7 Non-thinking Apr 2025 Official leaderboard
13 Gemini 2.0 Flash-Lite 16.6 Non-thinking Feb 2025 Official leaderboard
14 GPT-5 nano 16.4 Thinking Aug 2025 Official leaderboard
15 Command R (03-2024) 16.4 Non-thinking Mar 2024 Official leaderboard
16 Qwen3-30B-A3B 16.4 Non-thinking Apr 2025 Official leaderboard
17 Qwen3-14B 16.1 Non-thinking Apr 2025 Official leaderboard
18 Llama 4 Scout 15.7 Non-thinking Apr 2025 Official leaderboard
19 Qwen3-8B 15.6 Non-thinking Apr 2025 Official leaderboard
20 gpt-oss-120b 15.4 Thinking Aug 2025 Official leaderboard
21 Qwen3-4B 13.9 Non-thinking Apr 2025 Official leaderboard
22 Gemini 2.5 Flash Sep 2025 13.7 Thinking Sep 2025 Official leaderboard
23 gpt-oss-20b 13.3 Thinking Aug 2025 Official leaderboard
24 Mistral Small 3.1 13.2 Non-thinking Mar 2025 Official leaderboard
25 Gemma 3 12B 12.0 Non-thinking Mar 2025 Official leaderboard
26 Aya Expanse 32B 12.0 Non-thinking Oct 2024 Official leaderboard
27 Gemini 2.5 Pro Jun 2025 11.5 Thinking Jun 2025 Official leaderboard
28 Cohere Command A 11.4 Non-thinking Mar 2025 Official leaderboard
29 Llama 3.2 3B 11.3 Non-thinking Sep 2024 Official leaderboard
30 deepseek-r1-0528-qwen3-8b 11.2 Thinking May 2025 Official leaderboard
31 Claude 3.5 Haiku 10.9 Non-thinking Oct 2024 Official leaderboard
32 Gemma 3 27B 10.7 Non-thinking Mar 2025 Official leaderboard
33 Gemma 3 4B 9.50 Non-thinking Mar 2025 Official leaderboard
34 Gemma 3n E4B 9.40 Non-thinking Jun 2025 Official leaderboard
35 Llama 3.2 1B 8.23 Non-thinking Sep 2024 Official leaderboard
36 Gemma 3 1B 6.83 Non-thinking Mar 2025 Official leaderboard
37 Gemma 3 270M 5.10 Non-thinking Aug 2025 Official leaderboard

Rankings by part

We read 31 parts from the benchmark’s own leaderboard in 7 groups: question-generation persianqa, translation-en2fa en2fa, summarization samsum-fa, translation-fa2en fa2en, translation-ar2fa ar2fa, summarization pnsummary and translation-fa2ar fa2ar. Each ranks models on one part of the test. Llama 3.3 70B Instruct leads 11 of them, and 8 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

rougeL precision

37 models we track have a score here.

  1. 1 Gemini 2.0 Flash Feb 2025 ↑7 31.4
  2. 2 Gemini 2.5 Flash Sep 2025 ↑20 31.2
  3. 3 Gemini 2.5 Pro Jun 2025 ↑24 29.8
  4. 4 Gemini 2.0 Flash-Lite ↑9 27.8
  5. 5 GPT-5 mini ↑1 26.0
  6. 6 GPT-4.1 ↓4 25.8
  7. 7 Gemma 3 12B ↑18 25.7
  8. 8 Llama 3.3 70B Instruct ↓7 24.5
  9. 9 GPT-4.1 mini ↓6 24.4
  10. 10 Qwen3-32B ↑1 22.8

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…
18.2
GPT-5 nano
Thinking minimal rea…
16.4
gpt-oss-120b
Thinking low reasoni…
15.4
gpt-oss-20b
Thinking
13.3

Source: Official Persian NLG leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Llama 3.3 70B Instruct Non-thinking 20.1
GPT-4.1 Non-thinking 19.5
GPT-4.1 mini Non-thinking 19.0
GPT-4o Mar 2025 Non-thinking 19.0
Command R+ Non-thinking 18.8
GPT-5 mini Thinking 18.2
GPT-4o mini Non-thinking 18.1
Gemini 2.0 Flash Feb 2025 Non-thinking 17.8
Claude 3.7 Sonnet Non-thinking 17.8
o3 Thinking 17.6
Qwen3-32B Non-thinking 16.8
GPT-4.1 nano Non-thinking 16.7
Gemini 2.0 Flash-Lite Non-thinking 16.6
GPT-5 nano Thinking 16.4
Command R (03-2024) Non-thinking 16.4
Qwen3-30B-A3B Non-thinking 16.4
GPT-5 mini minimal reasoning 16.3
Qwen3-14B Non-thinking 16.1
Llama 4 Scout Non-thinking 15.7
Qwen3-8B Non-thinking 15.6
gpt-oss-120b Thinking 15.4
gpt-oss-120b low reasoning 14.3
GPT-5 nano minimal reasoning 14.2
Qwen3-4B Non-thinking 13.9
Gemini 2.5 Flash Sep 2025 Thinking 13.7
gpt-oss-20b Thinking 13.3
gpt-oss-20b low reasoning 13.2
Mistral Small 3.1 Non-thinking 13.2
Gemma 3 12B Non-thinking 12.0
Aya Expanse 32B Non-thinking 12.0
Gemini 2.5 Pro Jun 2025 Thinking 11.5
Cohere Command A Non-thinking 11.4
Llama 3.2 3B Non-thinking 11.3
deepseek-r1-0528-qwen3-8b Thinking 11.2
Claude 3.5 Haiku Non-thinking 10.9
Gemma 3 27B Non-thinking 10.7
Gemma 3 4B Non-thinking 9.50
Gemma 3n E4B Non-thinking 9.40
Llama 3.2 1B Non-thinking 8.23
Gemma 3 1B Non-thinking 6.83
Gemma 3 270M Non-thinking 5.10

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…
18.2
GPT-5 nano
Thinking minimal rea…
16.4
gpt-oss-120b
Thinking low reasoni…
15.4
gpt-oss-20b
Thinking
13.3

Source: Official Persian NLG leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Llama 3.3 70B Instruct Non-thinking 20.1
GPT-4.1 Non-thinking 19.5
GPT-4.1 mini Non-thinking 19.0
GPT-4o Mar 2025 Non-thinking 19.0
Command R+ Non-thinking 18.8
GPT-5 mini Thinking 18.2
GPT-4o mini Non-thinking 18.1
Gemini 2.0 Flash Feb 2025 Non-thinking 17.8
Claude 3.7 Sonnet Non-thinking 17.8
o3 Thinking 17.6
Qwen3-32B Non-thinking 16.8
GPT-4.1 nano Non-thinking 16.7
Gemini 2.0 Flash-Lite Non-thinking 16.6
GPT-5 nano Thinking 16.4
Command R (03-2024) Non-thinking 16.4
Qwen3-30B-A3B Non-thinking 16.4
GPT-5 mini minimal reasoning 16.3
Qwen3-14B Non-thinking 16.1
Llama 4 Scout Non-thinking 15.7
Qwen3-8B Non-thinking 15.6
gpt-oss-120b Thinking 15.4
gpt-oss-120b low reasoning 14.3
GPT-5 nano minimal reasoning 14.2
Qwen3-4B Non-thinking 13.9
Gemini 2.5 Flash Sep 2025 Thinking 13.7
gpt-oss-20b Thinking 13.3
gpt-oss-20b low reasoning 13.2
Mistral Small 3.1 Non-thinking 13.2
Gemma 3 12B Non-thinking 12.0
Aya Expanse 32B Non-thinking 12.0
Gemini 2.5 Pro Jun 2025 Thinking 11.5
Cohere Command A Non-thinking 11.4
Llama 3.2 3B Non-thinking 11.3
deepseek-r1-0528-qwen3-8b Thinking 11.2
Claude 3.5 Haiku Non-thinking 10.9
Gemma 3 27B Non-thinking 10.7
Gemma 3 4B Non-thinking 9.50
Gemma 3n E4B Non-thinking 9.40
Llama 3.2 1B Non-thinking 8.23
Gemma 3 1B Non-thinking 6.83
Gemma 3 270M Non-thinking 5.10

Number one over time

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

Meta
Own leaderboard Not tested since Sep 2025

Source: Official Persian NLG 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
5.00 10.0 15.0 20.0 25.0 2024 2025 5.00 10.0 15.0 20.0 25.0 2024 2025

Source: Official Persian NLG 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

8 of 9 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 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (41 entries).
  • Pass Models scored 37 models we track, plus 4 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 plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on mcinext-mizan-llm-leaderboard.hf.space: 41 rows for models we track and 4 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