Benchmarks › CritPt
CritPt
By CritPt research collaboration (UIUC, Argonne National Laboratory and others) ↗Results by Artificial Analysis ↗Since 2025Paper ↗
Solve unpublished research-level physics problems
32.9top score · Muse Wonder Full
321models ranked
215open-weights models
Pass@1higher is better
Oct 20, 2026last updated
How the test works
1
Task
70 tasks
- What the model doesSolve unpublished research-level physics problems
- Kinds of tasksPhysics research problems
- In → outText → Python functions, symbolic expressions, numerical answers
2
Run
5 runs per task
- Token budgetNon-reasoning: up to 16,384 output tokens, adjusted to model limits; reasoning: model-specific maximum
- Temperature0 for non-reasoning; 0.6 for reasoning, unless the model lab recommends otherwise
3
Score
Pass@1 (solved on the first try)
- Graded byOfficial CritPt grading server
- Better isHigher
Sources: artificialanalysis.ai
CritPt leaderboardtop 10 of 321
Full leaderboard with filters →| Model | CritPt | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
32.9 | — | — | 1M | |||
32.3max | $1.99 | $4 / $20 | 114 | 1.05M | ||
31.7max | $5.98 | $4 / $20 | 96 | 1M | ||
31.7max | $3.26 | $10 / $50 | 69 | 1.05M | ||
31.7max | $0.72 | $2 / $10 | 52 | 1.05M | ||
31.4max | $5.46 | $2 / $10 | 137 | 1M | ||
31.1xhigh | $5.98 | $10 / $50 | 64 | 1M | ||
30.9max | $1.04 | $2 / $10 | 128 | 1.05M | ||
30.6xhigh | — | $30 / $180 | — | 1.05M | ||
30xhigh | — | $30 / $180 | — | 1.05M |
CritPt questions
What is CritPt?›
Solve unpublished research-level physics problems. It has 70 tasks. Created by CritPt research collaboration (UIUC, Argonne National Laboratory and others).
Which model is best on CritPt?›
Muse Wonder Full leads CritPt with 32.9, ahead of GPT-5.6 Sol (32.3) and Claude Opus 5.5 (31.7).
What is the best open-weights model on CritPt?›
MiMo v2.6 Pro with 26.6, then DeepSeek V4.1 Flash 1017 (24).
How is CritPt scored?›
Pass@1 (solved on the first try); official critpt grading server; 5 runs per task per task.