MMLU-SR
Models answer MMLU questions after key terms are replaced with synonyms. The benchmark tests whether models recognize the same concept under different wording.
- Made by
- Wentian Wang and colleagues
- Released
- June 2024
- Kind
- Exam questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- MMLU
- In views
- Biology and medicine, Law and finance
- Strongest model tested
- GPT-4o mini #178 on the capability index
- Last model added
- July 2024
- Scores refreshed
- November 20th, 2026
GPT-4o (snapshot unspecified) tops the leaderboard with 74.8% GPT-4o mini leads with 65% among models released by July 2024
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline averages five-shot accuracy for the three symbol-replacement variants. 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-4o mini, at 65%. This view shows only models we track, each at its best.
The newest model with a score here came out in July 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 GPT-4o (snapshot unspecified) 5-shot 74.8%
- 2 Gemini-1.5-pro (snapshot unspecified) 5-shot 73.9%
- 3 GPT-4o mini 5-shot 65%
- 4
Llama 3-70B 5-shot 63%
- 5
Gemini 1.0 Pro 5-shot 59.2%
- 6 GPT-3.5-turbo (snapshot unspecified) 5-shot 53.2%
- 7
Llama 3-8B 5-shot 52.3%
Source: The MMLU-SR paper
- 1 GPT-4o mini 5-shot 65%
- 2
Llama 3-70B 5-shot 63%
- 3
Gemini 1.0 Pro 5-shot 59.2%
- 4
Llama 3-8B 5-shot 52.3%
Source: The MMLU-SR paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o (snapshot unspecified) | 74.8% | 5-shot | Official leaderboard | |
| 2 | Gemini-1.5-pro (snapshot unspecified) | 73.9% | 5-shot | Official leaderboard | |
| 3 | GPT-4o mini | 65% | 5-shot | Jul 2024 | Official leaderboard |
| 4 | | 63% | 5-shot | Apr 2024 | Official leaderboard |
| 5 | | 59.2% | 5-shot | Dec 2023 | Official leaderboard |
| 6 | GPT-3.5-turbo (snapshot unspecified) | 53.2% | 5-shot | Official leaderboard | |
| 7 | | 52.3% | 5-shot | Apr 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o mini | 65% | 5-shot | Jul 2024 | Official leaderboard |
| 2 | | 63% | 5-shot | Apr 2024 | Official leaderboard |
| 3 | | 59.2% | 5-shot | Dec 2023 | Official leaderboard |
| 4 | | 52.3% | 5-shot | Apr 2024 | Official leaderboard |
Rankings by part
We read 15 parts from the benchmark’s paper in 3 groups: question only, answer only and question and answer. Each ranks models on one part of the test. GPT-4o mini leads 9 of them, and Llama 3-70B leads the other 6. Arrows show how far a model moved from its overall rank among the models we track.
Humanities
4 models we track have a score here.
- 1 GPT-4o mini 74.4%
- 2
Gemini 1.0 Pro ↑1 68.7%
- 3
Llama 3-70B ↓1 63.5%
- 4
Llama 3-8B 54.6%
Social Sciences
4 models we track have a score here.
- 1
Llama 3-70B ↑1 81.2%
- 2 GPT-4o mini ↓1 79.2%
- 3
Gemini 1.0 Pro 74.4%
- 4
Llama 3-8B 68.5%
STEM
4 models we track have a score here.
- 1
Llama 3-70B ↑1 63.1%
- 2 GPT-4o mini ↓1 62.1%
- 3
Gemini 1.0 Pro 53.9%
- 4
Llama 3-8B 50.7%
Other
4 models we track have a score here.
- 1
Llama 3-70B ↑1 77%
- 2 GPT-4o mini ↓1 72.4%
- 3
Llama 3-8B ↑1 66.8%
- 4
Gemini 1.0 Pro ↓1 65.8%
Average
4 models we track have a score here.
- 1
Llama 3-70B ↑1 71.2%
- 2 GPT-4o mini ↓1 71%
- 3
Gemini 1.0 Pro 64.5%
- 4
Llama 3-8B 59.5%
Humanities
4 models we track have a score here.
- 1 GPT-4o mini 65.9%
- 2
Gemini 1.0 Pro ↑1 61.9%
- 3
Llama 3-70B ↓1 53.9%
- 4
Llama 3-8B 45.5%
Social Sciences
4 models we track have a score here.
- 1 GPT-4o mini 73.8%
- 2
Llama 3-70B 68.3%
- 3
Gemini 1.0 Pro 67%
- 4
Llama 3-8B 59.9%
STEM
4 models we track have a score here.
- 1 GPT-4o mini 60.2%
- 2
Llama 3-70B 56.5%
- 3
Gemini 1.0 Pro 50.4%
- 4
Llama 3-8B 46%
Other
4 models we track have a score here.
- 1 GPT-4o mini 65.1%
- 2
Llama 3-70B 62.2%
- 3
Gemini 1.0 Pro 59.1%
- 4
Llama 3-8B 55.7%
Average
4 models we track have a score here.
- 1 GPT-4o mini 65.5%
- 2
Llama 3-70B 60.2%
- 3
Gemini 1.0 Pro 58.6%
- 4
Llama 3-8B 51%
Humanities
4 models we track have a score here.
- 1 GPT-4o mini 58.8%
- 2
Gemini 1.0 Pro ↑1 58.2%
- 3
Llama 3-70B ↓1 52.3%
- 4
Llama 3-8B 42.1%
Social Sciences
4 models we track have a score here.
- 1 GPT-4o mini 66.6%
- 2
Llama 3-70B 65.3%
- 3
Gemini 1.0 Pro 62.2%
- 4
Llama 3-8B 53.8%
STEM
4 models we track have a score here.
- 1
Llama 3-70B ↑1 53.6%
- 2 GPT-4o mini ↓1 53.1%
- 3
Gemini 1.0 Pro 47.2%
- 4
Llama 3-8B 42.4%
Other
4 models we track have a score here.
- 1
Llama 3-70B ↑1 59.1%
- 2 GPT-4o mini ↓1 58.5%
- 3
Gemini 1.0 Pro 54.4%
- 4
Llama 3-8B 49.9%
Average
4 models we track have a score here.
- 1 GPT-4o mini 58.5%
- 2
Llama 3-70B 57.6%
- 3
Gemini 1.0 Pro 54.6%
- 4
Llama 3-8B 46.5%
2 models have held number one in the benchmark’s paper since its release in June 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MMLU-SR paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MMLU-SR 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 Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects.
Intern-S2-397B 89.8%
- MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy.
Seed1.8 92.3%
- SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy.
Qwen 3.6 Max (Preview) 73.9%
- WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1.
Tencent Hy4 preview 83.9%
- BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly.
Gemma 4 31B IT 74.4%
- HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in July 2024, for GPT-4o mini. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (7 entries).
- Pass Models scored 4 models we track, plus 3 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in July 2024, for GPT-4o mini.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4o mini (#178).
- Pass Headroom The best is 65%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 4 rows for models we track and 3 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.