MedPhysBench
Models calculate doses and check medical physics worksheets for planted errors. A grader checks the units and accepts values within a specified tolerance.
This evaluation tests calculations and error detection. MedPhysBench Knowledge Questions uses board-style multiple-choice questions.
- Made by
- Omar Nusrat
- Released
- November 2026
- Kind
- Medicine
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Subject
- Physical sciences, Clinical medicine
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Occupations
- Life, physical and social science
- In views
- Biology and medicine
- 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 100% among models released by September 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. The source ranks models by calculation accuracy and gives separate detection and false alarm rates. Where a model ran with several settings or harnesses, we keep its best.
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
- GPT-6.1 Terra
- and 10 more
- 1 GPT-6 Astra Stratified half, no tools 100%
- 2 GPT-6.1 Sol Full test, no tools 99.7%
- 3
Gemini 4 Argon Full test, no tools 99.5%
- 4 Claude Opus 5.5 Full test, no tools 99.3%
- 5 Claude Sonnet 5.5 Full test, no tools 93.7%
- 6 GPT-6 Luna Full test, no tools 93.5%
- 7 Claude Fable 5.1 Stratified half, no tools 90%
- 8
DeepSeek-V4-Pro Full test, no tools 89%
- 9 gpt-oss-120b Full test, no tools 85.7%
- 10
Mistral Large 3 Full test, no tools 84.8%
- 11 gpt-oss-20b Full test, no tools 76.5%
- 12
Gemma 3 27B Full test, no tools 33.3%
Source: Official MedPhysBench leaderboard
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 100% 98.8 to 100 | Stratified half, no tools | Sep 2026 | Official leaderboard |
| 2 | GPT-6.1 Sol | 99.7% 98.8 to 100 | Full test, no tools | Sep 2026 | Official leaderboard |
| 3 | | 99.5% 98.5 to 99.9 | Full test, no tools | Sep 2026 | Official leaderboard |
| 4 | Claude Opus 5.5 | 99.3% 98.3 to 99.8 | Full test, no tools | Sep 2026 | Official leaderboard |
| 5 | Claude Sonnet 5.5 | 93.7% 91.4 to 95.5 | Full test, no tools | Sep 2026 | Official leaderboard |
| 6 | GPT-6 Luna | 93.5% 91.2 to 95.3 | Full test, no tools | Sep 2026 | Official leaderboard |
| 7 | Claude Fable 5.1 | 90% 86 to 93.2 | Stratified half, no tools | Sep 2026 | Official leaderboard |
| 8 | | 89% 86.2 to 91.4 | Full test, no tools | Apr 2026 | Official leaderboard |
| 9 | gpt-oss-120b | 85.7% 82.6 to 88.4 | Full test, no tools | Aug 2025 | Official leaderboard |
| 10 | | 84.8% 81.7 to 87.6 | Full test, no tools | Dec 2025 | Official leaderboard |
| 11 | gpt-oss-20b | 76.5% 72.9 to 79.8 | Full test, no tools | Aug 2025 | Official leaderboard |
| 12 | | 33.3% 29.6 to 37.3 | Full test, no tools | Mar 2025 | Official leaderboard |
Rankings by part
We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Worksheet check detection, Result check detection and Worksheet check false alarms. Claude Fable 5.1 leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Worksheet check detection
12 models we track have a score here.
- 1 Claude Fable 5.1 ↑6 100%
- 2 GPT-6 Astra ↓1 100%
- 3
Gemini 4 Argon 100%
- 4 Claude Opus 5.5 99%
- 5 Claude Sonnet 5.5 99%
- 6 GPT-6.1 Sol ↓4 99%
- 7
Mistral Large 3 ↑3 96%
- 8
DeepSeek-V4-Pro 95%
- 9 GPT-6 Luna ↓3 90%
- 10 gpt-oss-120b ↓1 83%
Result check detection
12 models we track have a score here.
- 1
Gemini 4 Argon ↑2 100%
- 2 Claude Opus 5.5 ↑2 99%
- 3 Claude Sonnet 5.5 ↑2 99%
- 4 gpt-oss-120b ↑5 97%
- 5 Claude Fable 5.1 ↑2 97%
- 6 GPT-6 Astra ↓5 97%
- 7 GPT-6.1 Sol ↓5 97%
- 8 gpt-oss-20b ↑3 96%
- 9
Gemma 3 27B ↑3 95%
- 10 GPT-6 Luna ↓4 95%
Worksheet check false alarms
12 models we track have a score here.
- 1 Claude Fable 5.1 ↑6 0%
- 2 GPT-6.1 Sol 0%
- 3 GPT-6 Astra ↓2 1.00%
- 4 Claude Opus 5.5 1.00%
- 5
Gemini 4 Argon ↓2 1.00%
- 6 Claude Sonnet 5.5 ↓1 2.00%
- 7 GPT-6 Luna ↓1 3.00%
- 8 gpt-oss-120b ↑1 5.00%
- 9 gpt-oss-20b ↑2 5.00%
- 10
DeepSeek-V4-Pro ↓2 42%
Result check false alarms
12 models we track have a score here.
- 1 GPT-6 Astra 0%
- 2 GPT-6.1 Sol 0%
- 3 Claude Fable 5.1 ↑4 1.00%
- 4
Gemini 4 Argon ↓1 2.00%
- 5 Claude Opus 5.5 ↓1 4.00%
- 6 Claude Sonnet 5.5 ↓1 11%
- 7 GPT-6 Luna ↓1 12%
- 8 gpt-oss-120b ↑1 25%
- 9
DeepSeek-V4-Pro ↓1 36%
- 10 gpt-oss-20b ↑1 39%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official MedPhysBench leaderboard
Gpt-oss-120b on the official leaderboard passed 85% in August 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MedPhysBench 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 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (12 entries).
- Pass Models scored 12 models we track.
- Pass Newest score The newest is from September 2026, for Gemini 4 Argon.
- Pass Frontier models 6 of the top 10 have a score, led by GPT-6 Astra (#4).
- Note Headroom Saturated. The best is 100%, past 85% since August 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 12 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.