MMLU-ProX
MMLU-ProX - multilingual extended MMLU benchmark, average across en/de/fr/es/it/ja/zh/hi/pt/ko.
Category
Multilingual
Score type
percent
Scale
100%
higher is better
Models tested
14
What is MMLU-ProX?
MMLU-ProX is a benchmark for performance in languages other than English. The official description says: "MMLU-ProX - multilingual extended MMLU benchmark, average across en/de/fr/es/it/ja/zh/hi/pt/ko". Below we describe in simple words what benchmarks of this type generally examine.
What does it test?
MMLU-ProX measures performance in languages other than english. In detail, it checks:
- understanding and answering in many different languages
- translation quality between language pairs
- keeping math and reasoning quality when the question is not in English
- treating low-resource languages fairly instead of defaulting to English
How to read the score
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.
Official methodology note: MMLU-ProX - multilingual extended MMLU benchmark, average across en/de/fr/es/it/ja/zh/hi/pt/ko.
Frequently asked questions
What is the MMLU-ProX benchmark?
MMLU-ProX is a benchmark for performance in languages other than English. The official description says: 'MMLU-ProX - multilingual extended MMLU benchmark, average across en/de/fr/es/it/ja/zh/hi/pt/ko'. Below we describe in simple words what benchmarks of this type generally examine.
What does MMLU-ProX test?
MMLU-ProX mainly measures performance in languages other than English. Concretely, it checks: understanding and answering in many different languages; translation quality between language pairs; keeping math and reasoning quality when the question is not in English.
How is the MMLU-ProX score calculated?
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.
Where do I get MMLU-ProX data and scores?
The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/mmlu-prox/scores) or pull it directly from the Open LLM Radar API with any HTTP client.
Benchmark data & downloads
Where to get, inspect and reuse the data behind this benchmark.
Benchmark dataset
The question/task data itself — download and evaluate locally
Benchmark metadata (JSON)
This page's benchmark record from the Open LLM Radar API
All model scores (JSON download)
Full score table as JSON; paginate with limit/offset params
Open LLM Radar API docs
Swagger UI to explore and reuse all radar API endpoints
Model scores 14 models tested
| # | Model | Score | Date | Source | Verified |
|---|---|---|---|---|---|
| 1 |
DeepSeek-V4-Pro
DeepSeek
|
85.6%
|
04.06.2026 | Report | Unverified |
| 2 |
Kimi K2.6
Moonshot AI
|
85%
|
04.06.2026 | Report | Unverified |
| 3 |
DeepSeek-V4-Flash
DeepSeek
|
84.3%
|
04.06.2026 | Report | Unverified |
| 4 |
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
NVIDIA
|
83%
|
04.06.2026 | Report | Verified |
| 5 |
Qwen3.5-122B-A10B
Qwen Team (Alibaba Cloud)
|
82.2%
|
01.02.2026 | Report | Verified |
| 6 |
Qwen3.5-27B
Qwen Team (Alibaba Cloud)
|
82.2%
|
01.02.2026 | Report | Unverified |
| 7 |
Qwen3.5-35B-A3B
Qwen Team (Alibaba Cloud)
|
81%
|
01.02.2026 | Report | Unverified |
| 8 |
Qwen3-Next-80B-A3B-Thinking
Qwen
|
78.7%
|
10.09.2025 | Report | Verified |
| 9 |
Qwen3-235B-A22B
Qwen Team (Alibaba Cloud)
|
77.9%
|
01.02.2026 | Report | Unverified |
| 10 |
Qwen3-30B-A3B-Thinking-2507
Qwen
|
76.4%
|
31.07.2025 | Report | Verified |
| 11 |
Kolibri-1
Aleph Alpha
|
75.5%
|
03.10.2026 | Report | Verified |
| 12 |
Qwen3-30B-A3B
Qwen
|
73.1%
|
29.04.2025 | Report | Verified |
| 13 |
Qwen3.5-4B
Qwen
|
71.5%
|
01.02.2026 | Report | Verified |
| 14 |
Qwen3-VL-30B-A3B
Qwen
|
70.9%
|
04.10.2025 | Report | Verified |