MatSciBench
Models solve college materials science problems, including questions with figures. The authors report accuracy separately for text and image questions.
- Made by
- Junkai Zhang and colleagues
- Released
- October 2025
- Kind
- Science questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Subject
- Materials engineering
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Chemistry and materials
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- August 2025
- Scores refreshed
- November 20th, 2026
GPT-5 leads with 64% among models released by August 2025 GPT-5 leads with 64% among models released by August 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports two overall accuracies, so this is their plain mean for questions with and without images. 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 August 2025, 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-5 Base reasoning method 64%
- 2 Gemini-2.5-Pro (snapshot unspecified) Base reasoning method 62.9%
- 3 o4-mini Base reasoning method 57.8%
- 4
Llama 4 Maverick Base reasoning method 56.7%
- 5 Claude Sonnet 4 Base reasoning method 56.5%
- 6 GPT-4.1 Base reasoning method 54.3%
- 7 Claude 3.7 Sonnet Base reasoning method 51.8%
Source: The MatSciBench paper
- 1 GPT-5 Base reasoning method 64%
- 2 o4-mini Base reasoning method 57.8%
- 3
Llama 4 Maverick Base reasoning method 56.7%
- 4 Claude Sonnet 4 Base reasoning method 56.5%
- 5 GPT-4.1 Base reasoning method 54.3%
- 6 Claude 3.7 Sonnet Base reasoning method 51.8%
Source: The MatSciBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 64% | Base reasoning method | Aug 2025 | Official leaderboard |
| 2 | Gemini-2.5-Pro (snapshot unspecified) | 62.9% | Base reasoning method | Official leaderboard | |
| 3 | o4-mini | 57.8% | Base reasoning method | Apr 2025 | Official leaderboard |
| 4 | | 56.7% | Base reasoning method | Apr 2025 | Official leaderboard |
| 5 | Claude Sonnet 4 | 56.5% | Base reasoning method | May 2025 | Official leaderboard |
| 6 | GPT-4.1 | 54.3% | Base reasoning method | Apr 2025 | Official leaderboard |
| 7 | Claude 3.7 Sonnet | 51.8% | Base reasoning method | Feb 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 64% | Base reasoning method | Aug 2025 | Official leaderboard |
| 2 | o4-mini | 57.8% | Base reasoning method | Apr 2025 | Official leaderboard |
| 3 | | 56.7% | Base reasoning method | Apr 2025 | Official leaderboard |
| 4 | Claude Sonnet 4 | 56.5% | Base reasoning method | May 2025 | Official leaderboard |
| 5 | GPT-4.1 | 54.3% | Base reasoning method | Apr 2025 | Official leaderboard |
| 6 | Claude 3.7 Sonnet | 51.8% | Base reasoning method | Feb 2025 | Official leaderboard |
Rankings by part
We read 14 parts from the benchmark’s paper in 2 groups: questions without images and questions with images. Each ranks models on one part of the test. GPT-5 leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Fail. Mech.
6 models we track have a score here.
- 1 GPT-5 74.3%
- 2 o4-mini 71.3%
- 3 Claude Sonnet 4 ↑1 71.3%
- 4
Llama 4 Maverick ↓1 68.3%
- 5 GPT-4.1 66%
- 6 Claude 3.7 Sonnet 62.6%
Fund. Mech.
6 models we track have a score here.
- 1 GPT-5 74.6%
- 2 o4-mini 69.1%
- 3 GPT-4.1 ↑2 65.3%
- 4 Claude Sonnet 4 64.1%
- 5
Llama 4 Maverick ↓2 62%
- 6 Claude 3.7 Sonnet 58.4%
Materials
6 models we track have a score here.
- 1 GPT-5 76%
- 2 o4-mini 72%
- 3 GPT-4.1 ↑2 69.1%
- 4 Claude Sonnet 4 69.1%
- 5
Llama 4 Maverick ↓2 68.4%
- 6 Claude 3.7 Sonnet 62.1%
Processes
6 models we track have a score here.
- 1 GPT-5 74.8%
- 2 o4-mini 68.7%
- 3
Llama 4 Maverick 65.7%
- 4 Claude Sonnet 4 65.7%
- 5 GPT-4.1 57.6%
- 6 Claude 3.7 Sonnet 55.6%
Properties
6 models we track have a score here.
- 1 GPT-5 72.8%
- 2 o4-mini 70.9%
- 3
Llama 4 Maverick 67.3%
- 4 Claude Sonnet 4 67.3%
- 5 GPT-4.1 67%
- 6 Claude 3.7 Sonnet 61%
Structures
6 models we track have a score here.
- 1 GPT-5 76.6%
- 2 o4-mini 72%
- 3 GPT-4.1 ↑2 69.3%
- 4 Claude Sonnet 4 69%
- 5
Llama 4 Maverick ↓2 68.3%
- 6 Claude 3.7 Sonnet 63.1%
Overall
6 models we track have a score here.
- 1 GPT-5 75%
- 2 o4-mini 71.2%
- 3 GPT-4.1 ↑2 68.5%
- 4 Claude Sonnet 4 68.5%
- 5
Llama 4 Maverick ↓2 67.4%
- 6 Claude 3.7 Sonnet 62.5%
Fail. Mech.
6 models we track have a score here.
- 1 GPT-5 48.8%
- 2
Llama 4 Maverick ↑1 40.5%
- 3 Claude Sonnet 4 ↑1 35.7%
- 4 o4-mini ↓2 33.3%
- 5 Claude 3.7 Sonnet ↑1 28.6%
- 6 GPT-4.1 ↓1 25%
Fund. Mech.
6 models we track have a score here.
- 1 GPT-5 56.9%
- 2
Llama 4 Maverick ↑1 47.9%
- 3 Claude Sonnet 4 ↑1 47.9%
- 4 Claude 3.7 Sonnet ↑2 47.3%
- 5 o4-mini ↓3 47.3%
- 6 GPT-4.1 ↓1 46.7%
Materials
6 models we track have a score here.
- 1 GPT-5 55.2%
- 2
Llama 4 Maverick ↑1 48.1%
- 3 Claude Sonnet 4 ↑1 47%
- 4 o4-mini ↓2 45.1%
- 5 Claude 3.7 Sonnet ↑1 42.5%
- 6 GPT-4.1 ↓1 41.4%
Processes
6 models we track have a score here.
- 1 GPT-5 65.6%
- 2 o4-mini 59.4%
- 3 Claude Sonnet 4 ↑1 53.1%
- 4 Claude 3.7 Sonnet ↑2 43.8%
- 5 GPT-4.1 43.8%
- 6
Llama 4 Maverick ↓3 40.6%
Properties
6 models we track have a score here.
- 1 GPT-5 52.6%
- 2 o4-mini 42.6%
- 3 Claude Sonnet 4 ↑1 42.6%
- 4
Llama 4 Maverick ↓1 42.1%
- 5 Claude 3.7 Sonnet ↑1 36.8%
- 6 GPT-4.1 ↓1 34.5%
Structures
6 models we track have a score here.
- 1 GPT-5 53.9%
- 2
Llama 4 Maverick ↑1 46.8%
- 3 Claude Sonnet 4 ↑1 46.1%
- 4 o4-mini ↓2 44.7%
- 5 Claude 3.7 Sonnet ↑1 42.3%
- 6 GPT-4.1 ↓1 41.5%
Overall
6 models we track have a score here.
- 1 GPT-5 53%
- 2
Llama 4 Maverick ↑1 46%
- 3 o4-mini ↓1 44.4%
- 4 Claude Sonnet 4 44.4%
- 5 Claude 3.7 Sonnet ↑1 41%
- 6 GPT-4.1 ↓1 40%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MatSciBench 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 MatSciBench 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 Science and reasoning
- GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics.
Gemini 4 Carbon 97%
- Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
- BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
- FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics.
Intern-S2-397B 87%
- HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels.
Gemini 3 Pro 78.3%
- LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 15 months ago, in August 2025, for GPT-5. 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 6 models we track, plus 1 other entry on its leaderboard.
- Warning Newest score The newest came 15 months ago, in August 2025, for GPT-5.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 64%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 1 other entry 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.