SCI-AGENT-EXAM
Agents complete research tasks across 11 scientific fields. Programs check executable outputs, and a model judge grades written work.
- Made by
- Tianyun Zhong and colleagues
- Released
- October 2026
- Kind
- Research science
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Subject
- Computer and information sciences, Physical sciences, Chemical sciences, Earth and environmental sciences, Biological sciences
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal, Code sandbox
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents, Biology and medicine, Chemistry and materials
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
GPT-6 Astra leads with 80.9 among models released by September 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in September 2026, 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
- Claude Opus 5.5
- and 12 more
- 1 GPT-6 Astra 80.9
- 2 Claude Fable 5 67.3
- 3 Claude Opus 5 64.2
- 4
Qwen 3.8 Max 62.7
- 5 Kimi K3 62.6
- 6 GPT-5.6 Sol 59.6
- 7 Claude Opus 4.8 57.1
- 8 GLM-5.3 55.9
- 9 Grok 4.6 52.4
- 10
DeepSeek-V4-Pro 50.9
Source: Official SCI-AGENT-EXAM leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 80.9 | Sep 2026 | Official leaderboard | |
| 2 | Claude Fable 5 | 67.3 | Jun 2026 | Official leaderboard | |
| 3 | Claude Opus 5 | 64.2 | Jul 2026 | Official leaderboard | |
| 4 | | 62.7 | Aug 2026 | Official leaderboard | |
| 5 | Kimi K3 | 62.6 | Jul 2026 | Official leaderboard | |
| 6 | GPT-5.6 Sol | 59.6 | Jul 2026 | Official leaderboard | |
| 7 | Claude Opus 4.8 | 57.1 | May 2026 | Official leaderboard | |
| 8 | GLM-5.3 | 55.9 | Aug 2026 | Official leaderboard | |
| 9 | Grok 4.6 | 52.4 | Aug 2026 | Official leaderboard | |
| 10 | | 50.9 | Apr 2026 | Official leaderboard |
Rankings by part
We read 18 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Physical Science, Physics Simulation and Scientific Computation. GPT-6 Astra leads 17 of them, and Claude Fable 5 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
Physical Science
10 models we track have a score here.
- 1 GPT-6 Astra 82.8
- 2 Claude Opus 5 ↑1 68.9
- 3 Claude Fable 5 ↓1 55.4
- 4 Claude Opus 4.8 ↑3 53.2
- 5 GPT-5.6 Sol ↑1 52.8
- 6
Qwen 3.8 Max ↓2 50.1
- 7 Kimi K3 ↓2 49.9
- 8 GLM-5.3 47.5
- 9
DeepSeek-V4-Pro ↑1 40.7
- 10 Grok 4.6 ↓1 36.0
Physics Simulation
10 models we track have a score here.
- 1 GPT-6 Astra 81.7
- 2 Claude Opus 5 ↑1 66.3
- 3 Claude Fable 5 ↓1 64.5
- 4 Kimi K3 ↑1 62.5
- 5
Qwen 3.8 Max ↓1 61.3
- 6 Claude Opus 4.8 ↑1 59.7
- 7 Grok 4.6 ↑2 56.1
- 8 GLM-5.3 51.9
- 9 GPT-5.6 Sol ↓3 50.5
- 10
DeepSeek-V4-Pro 49.2
Scientific Computation
10 models we track have a score here.
- 1 GPT-6 Astra 82.3
- 2 Claude Opus 5 ↑1 67.7
- 3 GPT-5.6 Sol ↑3 63.7
- 4 Claude Fable 5 ↓2 61.8
- 5 Kimi K3 60.8
- 6
Qwen 3.8 Max ↓2 60.8
- 7 Claude Opus 4.8 60.1
- 8
DeepSeek-V4-Pro ↑2 58.2
- 9 Grok 4.6 58.2
- 10 GLM-5.3 ↓2 55.4
Data Science
10 models we track have a score here.
- 1 GPT-6 Astra 80.8
- 2 Claude Fable 5 75.8
- 3
Qwen 3.8 Max ↑1 65.4
- 4 Claude Opus 5 ↓1 64.4
- 5 Kimi K3 62.3
- 6 Claude Opus 4.8 ↑1 60.1
- 7 GLM-5.3 ↑1 58.5
- 8 GPT-5.6 Sol ↓2 57.1
- 9 Grok 4.6 54.9
- 10
DeepSeek-V4-Pro 49.3
Bioinformatics
10 models we track have a score here.
- 1 GPT-6 Astra 85.6
- 2
Qwen 3.8 Max ↑2 61.7
- 3 Kimi K3 ↑2 59.6
- 4 Claude Opus 5 ↓1 58.8
- 5 Grok 4.6 ↑4 58.3
- 6 GLM-5.3 ↑2 55.4
- 7 GPT-5.6 Sol ↓1 54.8
- 8 Claude Opus 4.8 ↓1 51.6
- 9 Claude Fable 5 ↓7 48.8
- 10
DeepSeek-V4-Pro 38.4
Biology
10 models we track have a score here.
- 1 Claude Fable 5 ↑1 71.4
- 2 GPT-6 Astra ↓1 69.8
- 3 Kimi K3 ↑2 66.7
- 4 GPT-5.6 Sol ↑2 64.2
- 5
DeepSeek-V4-Pro ↑5 59.7
- 6 Claude Opus 4.8 ↑1 59.7
- 7
Qwen 3.8 Max ↓3 59.7
- 8 Claude Opus 5 ↓5 59.6
- 9 GLM-5.3 ↓1 59.0
- 10 Grok 4.6 ↓1 47.6
Computer Science
10 models we track have a score here.
- 1 GPT-6 Astra 73.7
- 2 Claude Fable 5 69.5
- 3 Claude Opus 5 65.5
- 4 GLM-5.3 ↑4 56.7
- 5
Qwen 3.8 Max ↓1 56.4
- 6 Kimi K3 ↓1 55.8
- 7 GPT-5.6 Sol ↓1 54.9
- 8 Claude Opus 4.8 ↓1 47.6
- 9
DeepSeek-V4-Pro ↑1 47.5
- 10 Grok 4.6 ↓1 39.4
Materials Science
10 models we track have a score here.
- 1 GPT-6 Astra 86.5
- 2 Kimi K3 ↑3 81.4
- 3 Claude Fable 5 ↓1 75.5
- 4 GPT-5.6 Sol ↑2 71.4
- 5
Qwen 3.8 Max ↓1 68.9
- 6 Claude Opus 5 ↓3 66.5
- 7 Claude Opus 4.8 55.0
- 8
DeepSeek-V4-Pro ↑2 54.9
- 9 Grok 4.6 53.4
- 10 GLM-5.3 ↓2 49.5
Chemistry
10 models we track have a score here.
- 1 GPT-6 Astra 84.2
- 2 Claude Fable 5 77.2
- 3
Qwen 3.8 Max ↑1 71.9
- 4 Kimi K3 ↑1 65.8
- 5 GLM-5.3 ↑3 65.5
- 6 Grok 4.6 ↑3 65.1
- 7 Claude Opus 4.8 60.3
- 8 GPT-5.6 Sol ↓2 59.2
- 9
DeepSeek-V4-Pro ↑1 55.5
- 10 Claude Opus 5 ↓7 55.1
Mathematics
10 models we track have a score here.
- 1 GPT-6 Astra 78.1
- 2 Claude Opus 5 ↑1 69.8
- 3 Claude Fable 5 ↓1 69.5
- 4 GPT-5.6 Sol ↑2 65.3
- 5
Qwen 3.8 Max ↓1 61.3
- 6 Grok 4.6 ↑3 60.6
- 7 Kimi K3 ↓2 56.2
- 8 GLM-5.3 53.1
- 9 Claude Opus 4.8 ↓2 50.2
- 10
DeepSeek-V4-Pro 44.0
Engineering
10 models we track have a score here.
- 1 GPT-6 Astra 89.5
- 2 Claude Fable 5 79.8
- 3
Qwen 3.8 Max ↑1 78.7
- 4 Claude Opus 4.8 ↑3 71.2
- 5 Kimi K3 69.0
- 6 GLM-5.3 ↑2 66.5
- 7 Claude Opus 5 ↓4 65.4
- 8 GPT-5.6 Sol ↓2 61.7
- 9
DeepSeek-V4-Pro ↑1 60.4
- 10 Grok 4.6 ↓1 50.5
Code Generation
10 models we track have a score here.
- 1 GPT-6 Astra 79.0
- 2 Claude Fable 5 72.9
- 3 Claude Opus 5 71.9
- 4 GPT-5.6 Sol ↑2 64.1
- 5 Claude Opus 4.8 ↑2 63.0
- 6 Kimi K3 ↓1 58.7
- 7
Qwen 3.8 Max ↓3 58.2
- 8
DeepSeek-V4-Pro ↑2 52.9
- 9 Grok 4.6 45.8
- 10 GLM-5.3 ↓2 40.8
End-to-End Analysis
10 models we track have a score here.
- 1 GPT-6 Astra 83.9
- 2 Claude Opus 5 ↑1 67.3
- 3
Qwen 3.8 Max ↑1 64.1
- 4 Kimi K3 ↑1 61.5
- 5 GLM-5.3 ↑3 59.9
- 6 Claude Fable 5 ↓4 58.9
- 7 GPT-5.6 Sol ↓1 56.8
- 8 Claude Opus 4.8 ↓1 56.5
- 9 Grok 4.6 56.4
- 10
DeepSeek-V4-Pro 49.9
Error Diagnosis
10 models we track have a score here.
- 1 GPT-6 Astra 83.8
- 2 Claude Fable 5 68.0
- 3 Grok 4.6 ↑6 62.1
- 4 GPT-5.6 Sol ↑2 60.4
- 5 Claude Opus 5 ↓2 58.5
- 6
Qwen 3.8 Max ↓2 57.6
- 7 Kimi K3 ↓2 56.9
- 8 GLM-5.3 55.7
- 9
DeepSeek-V4-Pro ↑1 39.2
- 10 Claude Opus 4.8 ↓3 37.1
Hypothesis Validation
10 models we track have a score here.
- 1 GPT-6 Astra 77.3
- 2 Kimi K3 ↑3 61.9
- 3 Claude Fable 5 ↓1 60.0
- 4 Claude Opus 4.8 ↑3 58.0
- 5 Claude Opus 5 ↓2 57.9
- 6
Qwen 3.8 Max ↓2 57.2
- 7 GPT-5.6 Sol ↓1 56.0
- 8 GLM-5.3 55.6
- 9
DeepSeek-V4-Pro ↑1 49.6
- 10 Grok 4.6 ↓1 45.9
Literature Synthesis
10 models we track have a score here.
- 1 GPT-6 Astra 82.8
- 2
Qwen 3.8 Max ↑2 79.7
- 3 Claude Fable 5 ↓1 77.2
- 4 GLM-5.3 ↑4 74.1
- 5 Kimi K3 72.3
- 6 GPT-5.6 Sol 71.4
- 7
DeepSeek-V4-Pro ↑3 64.5
- 8 Claude Opus 4.8 ↓1 55.0
- 9 Claude Opus 5 ↓6 53.7
- 10 Grok 4.6 ↓1 29.1
Result Interpretation
10 models we track have a score here.
- 1 GPT-6 Astra 88.5
- 2 Claude Opus 5 ↑1 70.7
- 3 Claude Fable 5 ↓1 68.4
- 4
Qwen 3.8 Max 67.6
- 5 Kimi K3 66.7
- 6 Grok 4.6 ↑3 66.4
- 7 GPT-5.6 Sol ↓1 62.5
- 8 Claude Opus 4.8 ↓1 62.2
- 9 GLM-5.3 ↓1 54.6
- 10
DeepSeek-V4-Pro 51.8
Tool Orchestration
10 models we track have a score here.
- 1 GPT-6 Astra 77.2
- 2 Claude Fable 5 70.6
- 3 Kimi K3 ↑2 64.2
- 4 Claude Opus 5 ↓1 62.9
- 5
Qwen 3.8 Max ↓1 61.8
- 6 Claude Opus 4.8 ↑1 59.8
- 7 GLM-5.3 ↑1 57.9
- 8 Grok 4.6 ↑1 56.9
- 9 GPT-5.6 Sol ↓3 55.3
- 10
DeepSeek-V4-Pro 50.6
The best score, 80.9, costs $21 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official SCI-AGENT-EXAM leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-6 Astra | 80.9 | $21 | 28,583 | ||
| Claude Fable 5 | 67.3 | $7.67 | 51,087 | ||
| Claude Opus 5 | 64.2 | $12 | 67,797 | ||
| 62.7 | $22 | 114,405 | |||
| Kimi K3 | 62.6 | $21 | 63,987 | ||
| GPT-5.6 Sol | 59.6 | $13 | 21,638 | ||
| Claude Opus 4.8 | 57.1 | $5.13 | 57,566 | ||
| GLM-5.3 | 55.9 | $58 | 98,254 | ||
| Grok 4.6 | 52.4 | $13 | 100,347 | ||
| 50.9 | $32 | 95,669 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official SCI-AGENT-EXAM leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official SCI-AGENT-EXAM 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
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (10 entries).
- Pass Models scored 10 models we track.
- Pass Newest score The newest is from September 2026, for GPT-6 Astra.
- Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on skylenage-ai.github.io: 10 rows for models we track 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.