Benchmark Almanac

RoCS-MT v2

Yolu 1.00 tops its leaderboard GPT-4.1 7.00 leads the models we track
Not in model cards
Version 2

Models translate Reddit posts with revised spelling annotations and corrected references. The authors combine COMETKiwi and MetricX rankings for translations of the original text.

The authors rank 56 submissions, including GPT-4.1 and Claude 4. Bhojpuri and Icelandic have no human reference translations. Lower overall ranks place better.

Made by
Inria
Released
November 2025
Kind
Translation
Level
Everyday
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
Many languages
In and out
Text → Text
Interaction
Single turn
Items from
Real work and use
Run by
Anyone
Based on
RoCS-MT
In views
Many languages
Strongest model tested
GPT-4.1 #138 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
56
Entries in its paper, 10 models we track

Yolu tops the leaderboard with 1.00 GPT-4.1 leads with 7.00 among models released by April 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-4.1, at 7.00. This view shows only models we track, each at its best.

The newest model with a score here came out in April 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 Yolu 1.00
  2. 2 Shy-hunyuan-MT 2.00
  3. 3 CommandA-WMT 3.00
  4. 4 Laniqo 4.00
  5. 5 SalamandraTA 5.00
  6. 6 GemTrans 6.00
  7. 7 UvA-MT 7.00
  8. 8 GPT-4.1 7.00
  9. 9 ONLINE-B 9.00
  10. 10 TowerPlus 9B 10.0
  11. 11 TowerPlus 72B 11.0
  12. 12 SRPOL 12.0
  13. 13 IR-MultiagentMT 13.0
  14. 14 TranssionTranslate 14.0
  15. 15 CommandA 15.0
  16. 16 Erlendur 15.0
  17. 17 NNTSU 15.0
  18. 18 Algharb 18.0
  19. 19 Claude4 18.0
  20. 20 DeepSeek V3 (snapshot unspecified) 18.0
  21. 21 Gemini 2.5 Pro (snapshot unspecified) 18.0
  22. 22 In2x 18.0
  23. 23 Gemma 3 27B 23.0
  24. 24 Aya Expanse 32B 24.0
  25. 25 Aya Expanse 8B 25.0
  26. 26 KIKIS 26.0
  27. 27 EuroLLM 22B 27.0
  28. 28 Gemma 3 12B 28.0
  29. 29 Yandex 29.0
  30. 30 Systran 30.0
  31. 31 Kaze-MT 31.0
  32. 32 Llama 3.1-8B 31.0
  33. 33 KYUoM 33.0
  34. 34 ctpc_nlp 34.0
  35. 35 Wenyiil 35.0
  36. 36 Mistral Medium 36.0
  37. 37 Command R (03-2024) 37.0
  38. 38 Qwen3 235B 38.0
  39. 39 AMI 39.0
  40. 40 ONLINE-W 39.0
  41. 41 EuroLLM 9B 41.0
  42. 42 IRB-MT 42.0
  43. 43 Llama 4 Maverick 43.0
  44. 44 CUNI-MH-v2 44.0
  45. 45 bb88 45.0
  46. 46 NLLB 46.0
  47. 47 Mistral 7B v0.1 47.0
  48. 48 DLUT_GTCOM 48.0
  49. 49 CUNI-SFT 49.0
  50. 50 TranssionMT 50.0
  51. 51 Qwen 2.5 51.0
  52. 52 CGFOKUS 52.0
  53. 53 ONLINE-G 53.0
  54. 54 SH 54.0
  55. 55 COILD-BHO 55.0
  56. 56 CUNI-DocTransformer 55.0

Source: The RoCS-MT v2 paper

Source: The RoCS-MT v2 paper

#EntryScoreSettingDateSource
1 Yolu 1.00 Official leaderboard
2 Shy-hunyuan-MT 2.00 Official leaderboard
3 CommandA-WMT 3.00 Official leaderboard
4 Laniqo 4.00 Official leaderboard
5 SalamandraTA 5.00 Official leaderboard
6 GemTrans 6.00 Official leaderboard
7 UvA-MT 7.00 Official leaderboard
8 GPT-4.1 7.00 Apr 2025 Official leaderboard
9 ONLINE-B 9.00 Official leaderboard
10 TowerPlus 9B 10.0 Official leaderboard
11 TowerPlus 72B 11.0 Official leaderboard
12 SRPOL 12.0 Official leaderboard
13 IR-MultiagentMT 13.0 Official leaderboard
14 TranssionTranslate 14.0 Official leaderboard
15 CommandA 15.0 Official leaderboard
16 Erlendur 15.0 Official leaderboard
17 NNTSU 15.0 Official leaderboard
18 Algharb 18.0 Official leaderboard
19 Claude4 18.0 Official leaderboard
20 DeepSeek V3 (snapshot unspecified) 18.0 Official leaderboard
21 Gemini 2.5 Pro (snapshot unspecified) 18.0 Official leaderboard
22 In2x 18.0 Official leaderboard
23 Gemma 3 27B 23.0 Mar 2025 Official leaderboard
24 Aya Expanse 32B 24.0 Oct 2024 Official leaderboard
25 Aya Expanse 8B 25.0 Oct 2024 Official leaderboard
26 KIKIS 26.0 Official leaderboard
27 EuroLLM 22B 27.0 Official leaderboard
28 Gemma 3 12B 28.0 Mar 2025 Official leaderboard
29 Yandex 29.0 Official leaderboard
30 Systran 30.0 Official leaderboard
31 Kaze-MT 31.0 Official leaderboard
32 Llama 3.1-8B 31.0 Jul 2024 Official leaderboard
33 KYUoM 33.0 Official leaderboard
34 ctpc_nlp 34.0 Official leaderboard
35 Wenyiil 35.0 Official leaderboard
36 Mistral Medium 36.0 Dec 2023 Official leaderboard
37 Command R (03-2024) 37.0 Mar 2024 Official leaderboard
38 Qwen3 235B 38.0 Official leaderboard
39 AMI 39.0 Official leaderboard
40 ONLINE-W 39.0 Official leaderboard
41 EuroLLM 9B 41.0 Official leaderboard
42 IRB-MT 42.0 Official leaderboard
43 Llama 4 Maverick 43.0 Apr 2025 Official leaderboard
44 CUNI-MH-v2 44.0 Official leaderboard
45 bb88 45.0 Official leaderboard
46 NLLB 46.0 Official leaderboard
47 Mistral 7B v0.1 47.0 Sep 2023 Official leaderboard
48 DLUT_GTCOM 48.0 Official leaderboard
49 CUNI-SFT 49.0 Official leaderboard
50 TranssionMT 50.0 Official leaderboard
51 Qwen 2.5 51.0 Official leaderboard
52 CGFOKUS 52.0 Official leaderboard
53 ONLINE-G 53.0 Official leaderboard
54 SH 54.0 Official leaderboard
55 COILD-BHO 55.0 Official leaderboard
56 CUNI-DocTransformer 55.0 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-4.1 7.00 Apr 2025 Official leaderboard
2 Gemma 3 27B 23.0 Mar 2025 Official leaderboard
3 Aya Expanse 32B 24.0 Oct 2024 Official leaderboard
4 Aya Expanse 8B 25.0 Oct 2024 Official leaderboard
5 Gemma 3 12B 28.0 Mar 2025 Official leaderboard
6 Llama 3.1-8B 31.0 Jul 2024 Official leaderboard
7 Mistral Medium 36.0 Dec 2023 Official leaderboard
8 Command R (03-2024) 37.0 Mar 2024 Official leaderboard
9 Llama 4 Maverick 43.0 Apr 2025 Official leaderboard
10 Mistral 7B v0.1 47.0 Sep 2023 Official leaderboard

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 Nov 2025
GPT-4.1 7.00

Source: The RoCS-MT v2 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
0.00 10.0 20.0 30.0 40.0 50.0 2024 2025 0.00 10.0 20.0 30.0 40.0 50.0 2024 2025

Source: Official RoCS-MT v2 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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for GPT-4.1. 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 benchmark’s paper (56 entries).
  • Pass Models scored 10 models we track, plus 46 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for GPT-4.1.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4.1 (#138).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on aclanthology.org: 10 rows for models we track and 46 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