Benchmark Almanac

BenchMAX

Llama 3.3 70B 85.2 leads since February 2025
Not in model cards

Models follow instructions, write code, reason, use tools, translate, and retrieve information across 17 languages. Ten tasks use rule checks, model-judged wins, code tests, answer matching, or translation scores.

The suite translates and adapts existing evaluations, including IFEval, HumanEval+, LiveCodeBench, GPQA, and RULER. Human post-editing checks most translations. Component scores use different protocols and are not a single raw accuracy.

Released
February 2025
Kind
Reasoning in many languages
Level
University
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
Many languages
In and out
Text → Code, Text, Actions
Interaction
Agentic
Environment
APIs and tools
Refresh
One fixed set
Run by
Anyone
Strongest model tested
DeepSeek-V3 #155 on the capability index
Last model added
March 2025
Scores refreshed
November 20th, 2026

In the authors’ words

Previous multilingual benchmarks focus primarily on simple understanding tasks, but for large language models(LLMs), we emphasize proficiency in instruction following, reasoning, long context understanding, code generation, and so on. However, measuring these advanced capabilities across languages is underexplored. To address the disparity, we introduce BenchMAX, a multi-way multilingual evaluation benchmark that allows for fair comparisons of these important abilities across languages. To maintain high quality, three distinct native-speaking annotators independently annotate each sample within all tasks after the data was machine-translated from English into 16 other languages. Additionally, we present a novel translation challenge stemming from dataset construction. Extensive experiments on BenchMAX reveal varying effectiveness of core capabilities across languages, highlighting performance gaps that cannot be bridged by simply scaling up model size. BenchMAX serves as a comprehensive multilingual evaluation platform, providing a promising test bed to promote the development of multilingual language models. The dataset and code are publicly accessible.

Abstract of BenchMAX: A Comprehensive Multilingual Evaluation Suite for Large Language Models (arXiv:2502.07346, CC0)
24
Entries on its own leaderboard, 22 models we track

Llama 3.3 70B leads with 85.2 among models released by March 2025 Llama 3.3 70B leads with 85.2 among models released by March 2025

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. The authors report no overall score. The headline is rule-based instruction following, the first task in their main table. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in March 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 85.2
  2. 2 DeepSeek-V3 83.9
  3. 3 GPT-4o Aug 2024 80.8
  4. 4 Qwen2.5-72B 80.8
  5. 5 GPT-4o mini 79.1
  6. 6 Qwen2.5-32B 78.1
  7. 7 DeepSeek-R1-Distill-Llama-70B 78.0
  8. 8 Llama 3.1-70B 76.2
  9. 9 Gemma 3 4B 72.6
  10. 10 DeepSeek-R1-Distill-Qwen-32B 67.3
  11. 11 Qwen2.5-7B 65.9
  12. 12 Gemma 2 9B 63.0
  13. 13 Llama 3.1-8B 62.6
  14. 14 Gemma 2 27B 62.4
  15. 15 Aya Expanse 32B 61.9
  16. 16 InternLM2.5-20B-Chat 51.9
  17. 17 InternLM3-8B-Instruct 51.3
  18. 18 Aya Expanse 8B 51.2
  19. 19 Llama 3.2 3B 51.0
  20. 20 DeepSeek-R1-Distill-Llama 8B 49.7
  21. 21 Gemma 3 1B 47.1
  22. 22 DeepSeek-R1-Distill-Qwen-7B 46.7
  23. 23 InternLM2.5 7B 45.7
  24. 24 Llama 3.2 1B 34.0

Source: Official BenchMAX leaderboard

#EntryScoreSettingDateSource
1 Llama 3.3 70B 85.2 Dec 2024 Official leaderboard
2 DeepSeek-V3 83.9 Dec 2024 Official leaderboard
3 GPT-4o Aug 2024 80.8 Aug 2024 Official leaderboard
4 Qwen2.5-72B 80.8 Sep 2024 Official leaderboard
5 GPT-4o mini 79.1 Jul 2024 Official leaderboard
6 Qwen2.5-32B 78.1 Sep 2024 Official leaderboard
7 DeepSeek-R1-Distill-Llama-70B 78.0 Jan 2025 Official leaderboard
8 Llama 3.1-70B 76.2 Jul 2024 Official leaderboard
9 Gemma 3 4B 72.6 Mar 2025 Official leaderboard
10 DeepSeek-R1-Distill-Qwen-32B 67.3 Jan 2025 Official leaderboard
11 Qwen2.5-7B 65.9 Sep 2024 Official leaderboard
12 Gemma 2 9B 63.0 Jun 2024 Official leaderboard
13 Llama 3.1-8B 62.6 Jul 2024 Official leaderboard
14 Gemma 2 27B 62.4 Jun 2024 Official leaderboard
15 Aya Expanse 32B 61.9 Oct 2024 Official leaderboard
16 InternLM2.5-20B-Chat 51.9 Aug 2024 Official leaderboard
17 InternLM3-8B-Instruct 51.3 Jan 2025 Official leaderboard
18 Aya Expanse 8B 51.2 Oct 2024 Official leaderboard
19 Llama 3.2 3B 51.0 Sep 2024 Official leaderboard
20 DeepSeek-R1-Distill-Llama 8B 49.7 Official leaderboard
21 Gemma 3 1B 47.1 Mar 2025 Official leaderboard
22 DeepSeek-R1-Distill-Qwen-7B 46.7 Jan 2025 Official leaderboard
23 InternLM2.5 7B 45.7 Official leaderboard
24 Llama 3.2 1B 34.0 Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Llama 3.3 70B 85.2 Dec 2024 Official leaderboard
2 DeepSeek-V3 83.9 Dec 2024 Official leaderboard
3 GPT-4o Aug 2024 80.8 Aug 2024 Official leaderboard
4 Qwen2.5-72B 80.8 Sep 2024 Official leaderboard
5 GPT-4o mini 79.1 Jul 2024 Official leaderboard
6 Qwen2.5-32B 78.1 Sep 2024 Official leaderboard
7 DeepSeek-R1-Distill-Llama-70B 78.0 Jan 2025 Official leaderboard
8 Llama 3.1-70B 76.2 Jul 2024 Official leaderboard
9 Gemma 3 4B 72.6 Mar 2025 Official leaderboard
10 DeepSeek-R1-Distill-Qwen-32B 67.3 Jan 2025 Official leaderboard
11 Qwen2.5-7B 65.9 Sep 2024 Official leaderboard
12 Gemma 2 9B 63.0 Jun 2024 Official leaderboard
13 Llama 3.1-8B 62.6 Jul 2024 Official leaderboard
14 Gemma 2 27B 62.4 Jun 2024 Official leaderboard
15 Aya Expanse 32B 61.9 Oct 2024 Official leaderboard
16 InternLM2.5-20B-Chat 51.9 Aug 2024 Official leaderboard
17 InternLM3-8B-Instruct 51.3 Jan 2025 Official leaderboard
18 Aya Expanse 8B 51.2 Oct 2024 Official leaderboard
19 Llama 3.2 3B 51.0 Sep 2024 Official leaderboard
20 Gemma 3 1B 47.1 Mar 2025 Official leaderboard
21 DeepSeek-R1-Distill-Qwen-7B 46.7 Jan 2025 Official leaderboard
22 Llama 3.2 1B 34.0 Sep 2024 Official leaderboard

Rankings by part

We read 16 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Rule-based instruction following, Model-based instruction following and Function completion. DeepSeek-V3 leads 9 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.

Rule-based instruction following

22 models we track have a score here.

  1. 1 Llama 3.3 70B 85.2%
  2. 2 DeepSeek-V3 83.9%
  3. 3 GPT-4o Aug 2024 80.8%
  4. 4 Qwen2.5-72B 80.8%
  5. 5 GPT-4o mini 79.1%
  6. 6 Qwen2.5-32B 78.1%
  7. 7 DeepSeek-R1-Distill-Llama-70B 78%
  8. 8 Llama 3.1-70B 76.2%
  9. 9 Gemma 3 4B 72.6%
  10. 10 DeepSeek-R1-Distill-Qwen-32B 67.3%

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 Mar 2025

Source: Official BenchMAX 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
20.0 40.0 60.0 80.0 100.0 2024 2025 20.0 40.0 60.0 80.0 100.0 2024 2025

Source: Official BenchMAX 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 20 months ago, in March 2025, for Gemma 3 1B. 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 (24 entries).
  • Pass Models scored 22 models we track, plus 2 other entries on its leaderboard.
  • Warning Newest score The newest came 20 months ago, in March 2025, for Gemma 3 1B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is DeepSeek-V3 (#155).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 22 rows for models we track and 2 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