GradeBench
Models read a student's handwritten answer to a university STEM exam question, with the rubric and a reference solution, and pick the rubric items to award. Scored by exact match with the items the instructor chose, averaged over five runs.
464 tasks from 125 questions in 45 courses at more than 10 universities, taken from real exams graded on Pensive. The tasks are not released, and the site is the only source besides an X post. Two human graders agree on 92% of tasks. GradeBench-Diagram is a 258-task subset where students draw.
- Made by
- Pensive
- Released
- November 2026
- Kind
- Other professions
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- Kept private
- Measures
- Capability
- Subject
- Mathematics, Physical sciences, Electrical, electronic and information engineering, Education
- Language
- English
- Interaction
- Single turn
- Items from
- Exams and contests, Real work and use
- Refresh
- One fixed set
- Run by
- A host
- Occupations
- Educational instruction and library
- In views
- Vision
- Strongest model tested
- GPT-6.2 Astra #1 on the capability index
- Last model added
- November 2026
- Scores refreshed
- November 18th, 2026
Claude Fable 5.2 leads with 57.3%
Best score per model from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
- 57.3 Claude Fable 5.2
- 56.6
Gemini 4 Carbon
- 56.5
Gemini 4 Argon
- 56.3 GPT-6.2 Astra
- 55.5 GPT-6.2 Sol
- 53.6 GPT-6 Astra
- 53.5 Claude Opus 5.5
- 53.2 GPT-6.1 Sol
- 53.1 Claude Fable 5.1
- 51.9
Gemini 4 Hydrogen
- 51.5
Gemini 3.8 Flash
- 48.1
Muse Wonder
- 43.7 Claude Sonnet 5.5
- 41.4
Muse Spark 1.3
- 36.6 GLM-5.3-Flash
- 36.1 GPT-6 Luna
- 35.3
DeepSeek V4.1 Flash
- 35.3 Grok 4.7
- 1 Claude Fable 5.2 single-shot user prompt; medium reasoning or nearest supported level 57.3%
- 2
Gemini 4 Carbon single-shot user prompt; medium reasoning or nearest supported level 56.6%
- 3
Gemini 4 Argon single-shot user prompt; medium reasoning or nearest supported level 56.5%
- 4 GPT-6.2 Astra single-shot user prompt; medium reasoning or nearest supported level 56.3%
- 5 GPT-6.2 Sol single-shot user prompt; medium reasoning or nearest supported level 55.5%
- 6 GPT-6 Astra single-shot user prompt; medium reasoning or nearest supported level 53.6%
- 7 Claude Opus 5.5 single-shot user prompt; medium reasoning or nearest supported level 53.5%
- 8 GPT-6.1 Sol single-shot user prompt; medium reasoning or nearest supported level 53.2%
- 9 Claude Fable 5.1 single-shot user prompt; medium reasoning or nearest supported level 53.1%
- 10
Gemini 4 Hydrogen single-shot user prompt; medium reasoning or nearest supported level 51.9%
- 11
Gemini 3.8 Flash single-shot user prompt; medium reasoning or nearest supported level 51.5%
- 12
Muse Wonder single-shot user prompt; medium reasoning or nearest supported level 48.1%
- 13 Claude Sonnet 5.5 single-shot user prompt; medium reasoning or nearest supported level 43.7%
- 14
Muse Spark 1.3 single-shot user prompt; medium reasoning or nearest supported level 41.4%
- 15 GLM-5.3-Flash single-shot user prompt; medium reasoning or nearest supported level 36.6%
- 16 GPT-6 Luna single-shot user prompt; medium reasoning or nearest supported level 36.1%
- 17
DeepSeek V4.1 Flash single-shot user prompt; medium reasoning or nearest supported level 35.3%
- 18 Grok 4.7 single-shot user prompt; medium reasoning or nearest supported level 35.3%
Source: Official GradeBench leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | 57.3% | single-shot user prompt; medium reasoning or nearest supported level | Nov 2026 | Official leaderboard |
| 2 | | 56.6% | single-shot user prompt; medium reasoning or nearest supported level | Oct 2026 | Official leaderboard |
| 3 | | 56.5% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 4 | GPT-6.2 Astra | 56.3% | single-shot user prompt; medium reasoning or nearest supported level | Nov 2026 | Official leaderboard |
| 5 | GPT-6.2 Sol | 55.5% | single-shot user prompt; medium reasoning or nearest supported level | Nov 2026 | Official leaderboard |
| 6 | GPT-6 Astra | 53.6% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 7 | Claude Opus 5.5 | 53.5% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 8 | GPT-6.1 Sol | 53.2% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 9 | Claude Fable 5.1 | 53.1% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 10 | | 51.9% | single-shot user prompt; medium reasoning or nearest supported level | Nov 2026 | Official leaderboard |
| 11 | | 51.5% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 12 | | 48.1% | single-shot user prompt; medium reasoning or nearest supported level | Oct 2026 | Official leaderboard |
| 13 | Claude Sonnet 5.5 | 43.7% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 14 | | 41.4% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 15 | GLM-5.3-Flash | 36.6% | single-shot user prompt; medium reasoning or nearest supported level | Aug 2026 | Official leaderboard |
| 16 | GPT-6 Luna | 36.1% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 17 | | 35.3% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
| 18 | Grok 4.7 | 35.3% | single-shot user prompt; medium reasoning or nearest supported level | Sep 2026 | Official leaderboard |
Rankings by part
We read 16 parts from the benchmark’s own leaderboard in 3 groups: domain, modality and category. Each ranks models on one part of the test. Claude Fable 5.2 leads 8 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Engineering
16 models we track have a score here.
- 1 Claude Fable 5.2 52.8%
- 2 GPT-6 Astra ↑4 49.8%
- 3
Gemini 4 Carbon ↓1 49.7%
- 4
Gemini 4 Argon ↓1 49.1%
- 5 Claude Opus 5.5 ↑2 48.6%
- 6 GPT-6.1 Sol ↑2 48.1%
- 7 Claude Fable 5.1 ↑2 47.2%
- 8
Gemini 4 Hydrogen ↑2 44.7%
- 9
Gemini 3.8 Flash ↑2 44.6%
- 10
Muse Wonder ↑2 41%
Chemistry
16 models we track have a score here.
- 1 GPT-6 Astra ↑5 66%
- 2 GPT-6.1 Sol ↑6 64.8%
- 3
Gemini 4 Carbon ↓1 64.1%
- 4
Gemini 4 Argon ↓1 63.8%
- 5
Gemini 4 Hydrogen ↑5 61.6%
- 6 Claude Fable 5.2 ↓5 60.3%
- 7 Claude Opus 5.5 56.7%
- 8 Claude Fable 5.1 ↑1 56.2%
- 9
Gemini 3.8 Flash ↑2 53.6%
- 10
Muse Wonder ↑2 49.7%
Mathematics
16 models we track have a score here.
- 1
Gemini 4 Carbon ↑1 63.7%
- 2 Claude Fable 5.1 ↑7 61.7%
- 3
Gemini 3.8 Flash ↑8 60%
- 4 Claude Fable 5.2 ↓3 59.7%
- 5
Gemini 4 Argon ↓2 58%
- 6 Claude Opus 5.5 ↑1 57.7%
- 7
Gemini 4 Hydrogen ↑3 54.3%
- 8
Muse Wonder ↑4 51.7%
- 9 GPT-6.1 Sol ↓1 48.3%
- 10 GPT-6 Astra ↓4 48%
Physics
16 models we track have a score here.
- 1 Claude Fable 5.2 59%
- 2
Gemini 4 Argon ↑1 56.6%
- 3 Claude Opus 5.5 ↑4 54.9%
- 4
Gemini 4 Carbon ↓2 54.6%
- 5 Claude Sonnet 5.5 ↑8 51.5%
- 6
Muse Wonder ↑6 51.5%
- 7
Gemini 3.8 Flash ↑4 48.5%
- 8 GLM-5.3-Flash ↑7 47.8%
- 9 GPT-6.1 Sol ↓1 47.8%
- 10 Claude Fable 5.1 ↓1 47.5%
Statistics
16 models we track have a score here.
- 1 Claude Fable 5.1 ↑8 63.7%
- 2
DeepSeek V4.1 Flash ↑15 60.9%
- 3 GPT-6.1 Sol ↑5 60.5%
- 4 GPT-6 Astra ↑2 60%
- 5
Muse Wonder ↑7 57.2%
- 6 Claude Sonnet 5.5 ↑7 56.3%
- 7 Claude Opus 5.5 55.8%
- 8
Gemini 4 Argon ↓5 55.8%
- 9
Gemini 4 Hydrogen ↑1 54.9%
- 10 Claude Fable 5.2 ↓9 54%
Computer science
16 models we track have a score here.
- 1 Claude Fable 5.2 62.4%
- 2
Gemini 3.8 Flash ↑9 60%
- 3
Gemini 4 Argon 60%
- 4
Gemini 4 Carbon ↓2 59%
- 5
Gemini 4 Hydrogen ↑5 51.7%
- 6 Claude Opus 5.5 ↑1 51.2%
- 7 Claude Fable 5.1 ↑2 50.2%
- 8
Muse Wonder ↑4 49.8%
- 9
Muse Spark 1.3 ↑5 48.8%
- 10 GLM-5.3-Flash ↑5 47.3%
Drawings
16 models we track have a score here.
- 1 Claude Fable 5.2 58.8%
- 2
Gemini 4 Carbon 56.7%
- 3
Gemini 4 Argon 55.5%
- 4 GPT-6 Astra ↑2 54.3%
- 5 GPT-6.1 Sol ↑3 52.9%
- 6 Claude Opus 5.5 ↑1 51.4%
- 7 Claude Fable 5.1 ↑2 51.2%
- 8
Gemini 4 Hydrogen ↑2 51%
- 9
Gemini 3.8 Flash ↑2 47.6%
- 10
Muse Wonder ↑2 46.8%
Written
16 models we track have a score here.
- 1
Gemini 4 Argon ↑2 57.7%
- 2
Gemini 4 Carbon 56.4%
- 3
Gemini 3.8 Flash ↑8 56.3%
- 4 Claude Opus 5.5 ↑3 56.1%
- 5 Claude Fable 5.1 ↑4 55.6%
- 6 Claude Fable 5.2 ↓5 55.4%
- 7 GPT-6.1 Sol ↑1 53.5%
- 8
Gemini 4 Hydrogen ↑2 53.1%
- 9 GPT-6 Astra ↓3 52.7%
- 10 Claude Sonnet 5.5 ↑3 51.9%
Graphs and plots
16 models we track have a score here.
- 1 Claude Fable 5.2 49.6%
- 2
Gemini 3.8 Flash ↑9 48.1%
- 3
Gemini 4 Argon 47.2%
- 4 Claude Fable 5.1 ↑5 46.6%
- 5 Claude Opus 5.5 ↑2 45.7%
- 6
Gemini 4 Carbon ↓4 45.7%
- 7
Gemini 4 Hydrogen ↑3 44.8%
- 8 GPT-6 Astra ↓2 44.2%
- 9 GPT-6.1 Sol ↓1 43.9%
- 10 Claude Sonnet 5.5 ↑3 43%
Chemical structures
16 models we track have a score here.
- 1
Gemini 4 Carbon ↑1 60%
- 2 GPT-6.1 Sol ↑6 59.6%
- 3
Gemini 4 Hydrogen ↑7 59.6%
- 4 GPT-6 Astra ↑2 59.2%
- 5
Gemini 4 Argon ↓2 58.8%
- 6 Claude Fable 5.2 ↓5 58%
- 7 Claude Opus 5.5 50.4%
- 8
Gemini 3.8 Flash ↑3 46%
- 9 Claude Fable 5.1 41.2%
- 10
Muse Wonder ↑2 37.6%
Circuits and timing
16 models we track have a score here.
- 1
Gemini 4 Carbon ↑1 62.1%
- 2 Claude Fable 5.2 ↓1 62.1%
- 3 GPT-6 Astra ↑3 58.7%
- 4
Gemini 4 Argon ↓1 58.3%
- 5 Claude Fable 5.1 ↑4 55.7%
- 6 GPT-6.1 Sol ↑2 52.8%
- 7
Gemini 4 Hydrogen ↑3 51.1%
- 8 Claude Opus 5.5 ↓1 50.2%
- 9
Muse Spark 1.3 ↑5 48.1%
- 10
Muse Wonder ↑2 47.7%
Orbital and energy levels
16 models we track have a score here.
- 1
Gemini 4 Carbon ↑1 85.6%
- 2
Gemini 4 Argon ↑1 83.8%
- 3 GPT-6 Astra ↑3 82.5%
- 4 Claude Fable 5.1 ↑5 80.6%
- 5 Claude Fable 5.2 ↓4 79.4%
- 6 GPT-6.1 Sol ↑2 78.1%
- 7
Gemini 4 Hydrogen ↑3 75.6%
- 8 Claude Opus 5.5 ↓1 73.8%
- 9
Gemini 3.8 Flash ↑2 68.8%
- 10
Muse Spark 1.3 ↑4 68.8%
Trees and graphs
16 models we track have a score here.
- 1 Claude Fable 5.2 54.2%
- 2 Claude Fable 5.1 ↑7 46.5%
- 3
Gemini 4 Carbon ↓1 46.5%
- 4
Gemini 3.8 Flash ↑7 43.2%
- 5
Muse Wonder ↑7 43.2%
- 6 Claude Opus 5.5 ↑1 42.6%
- 7
Gemini 4 Argon ↓4 42.6%
- 8 GPT-6 Astra ↓2 41.9%
- 9 GPT-6.1 Sol ↓1 41.9%
- 10 Claude Sonnet 5.5 ↑3 41.3%
Free-body
16 models we track have a score here.
- 1 GPT-6.1 Sol ↑7 71.8%
- 2 GPT-6 Astra ↑4 70.6%
- 3 Claude Fable 5.2 ↓2 65.9%
- 4
Muse Wonder ↑8 64.7%
- 5 Claude Opus 5.5 ↑2 63.5%
- 6 Claude Fable 5.1 ↑3 58.8%
- 7
Gemini 3.8 Flash ↑4 57.6%
- 8
Gemini 4 Hydrogen ↑2 55.3%
- 9
Gemini 4 Argon ↓6 54.1%
- 10 GPT-6 Luna ↑6 51.8%
Ray diagrams
16 models we track have a score here.
- 1 Claude Fable 5.2 50%
- 2
Gemini 4 Carbon 42.9%
- 3 Claude Opus 5.5 ↑4 40%
- 4
Gemini 4 Argon ↓1 40%
- 5
Muse Spark 1.3 ↑9 32.9%
- 6 Claude Sonnet 5.5 ↑7 32.9%
- 7
Muse Wonder ↑5 32.9%
- 8 Claude Fable 5.1 ↑1 27.1%
- 9
DeepSeek V4.1 Flash ↑8 25.7%
- 10
Gemini 4 Hydrogen 25.7%
All drawings
16 models we track have a score here.
- 1 Claude Fable 5.2 58.8%
- 2
Gemini 4 Carbon 56.7%
- 3
Gemini 4 Argon 55.5%
- 4 GPT-6 Astra ↑2 54.3%
- 5 GPT-6.1 Sol ↑3 52.9%
- 6 Claude Opus 5.5 ↑1 51.4%
- 7 Claude Fable 5.1 ↑2 51.2%
- 8
Gemini 4 Hydrogen ↑2 51%
- 9
Gemini 3.8 Flash ↑2 47.6%
- 10
Muse Wonder ↑2 46.8%
The best score, 57.3%, costs $0.07 a task
Runs on the cost frontier score higher than any run that's cheaper. Gemini 4 Argon at single-shot user prompt; medium reasoning or nearest supported level comes within 0.8 points of the best for $0.009 a task, 7 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official GradeBench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Fable 5.2 | single-shot user prompt; medium reasoning or nearest supported level | 57.3% | $0.07 | 252 | |
| single-shot user prompt; medium reasoning or nearest supported level | 56.6% | $0.04 | 3,106 | ||
| single-shot user prompt; medium reasoning or nearest supported level | 56.5% | $0.009 | 291 | ||
| GPT-6.2 Astra | single-shot user prompt; medium reasoning or nearest supported level | 56.3% | $0.04 | 34 | |
| GPT-6.2 Sol | single-shot user prompt; medium reasoning or nearest supported level | 55.5% | $0.01 | 57 | |
| GPT-6 Astra | single-shot user prompt; medium reasoning or nearest supported level | 53.6% | $0.06 | 96 | |
| Claude Opus 5.5 | single-shot user prompt; medium reasoning or nearest supported level | 53.5% | $0.04 | 423 | |
| GPT-6.1 Sol | single-shot user prompt; medium reasoning or nearest supported level | 53.2% | $0.01 | 113 | |
| Claude Fable 5.1 | single-shot user prompt; medium reasoning or nearest supported level | 53.1% | $0.10 | 603 | |
| single-shot user prompt; medium reasoning or nearest supported level | 51.9% | $0.02 | 1,793 | ||
| single-shot user prompt; medium reasoning or nearest supported level | 51.5% | $0.01 | 1,966 | ||
| single-shot user prompt; medium reasoning or nearest supported level | 48.1% | $0.06 | 1,776 | ||
| Claude Sonnet 5.5 | single-shot user prompt; medium reasoning or nearest supported level | 43.7% | $0.02 | 392 | |
| single-shot user prompt; medium reasoning or nearest supported level | 41.4% | $0.02 | 3,189 | ||
| GLM-5.3-Flash | single-shot user prompt; medium reasoning or nearest supported level | 36.6% | $0.001 | 2,190 | |
| GPT-6 Luna | single-shot user prompt; medium reasoning or nearest supported level | 36.1% | $0.001 | 1,408 | |
| single-shot user prompt; medium reasoning or nearest supported level | 35.3% | $0.01 | 12,885 | ||
| Grok 4.7 | single-shot user prompt; medium reasoning or nearest supported level | 35.3% | $0.09 | 12,786 |
2 models have held number one on the official leaderboard since its release in November 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official GradeBench 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 GradeBench 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 Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
All 11 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access The tasks are private.
- Pass Confidence Checked against a primary source.
Scores 5 pass
- Pass Independent runs Scores from the official leaderboard (18 entries).
- Pass Models scored 18 models we track.
- Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
- Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
- Pass Headroom The best is 57.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on gradebench.com: 18 rows for models we track at the last read.
- Pass Last read November 18th, 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.