Updated daily at 05:00 UTC
Benchmark
Visual Physics Comprehension Test (VPCT)
Measures: 100 diagrams of a ball rolling down ramps into one of several buckets; the model predicts which bucket it lands in. Human volunteers scored 100%.
- Maintainer
- —
- Introduced
- 2025
- Status
- Active
- Saturation
- Near-ceiling
- Typical provenance
- Independent
- In scoring set
- Yes
Leaderboard
Independent and third-party results only.
| Model | Correct | Provenance | Source | Evaluated |
|---|---|---|---|---|
| Claude Sonnet 4 | 30.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Flash-Lite | 30.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude 3.5 Sonnet | 33.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude 3.5 Sonnet | 33.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Opus 4 | 33.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Opus 4.1 Anthropic · 16K effort | 33.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Sonnet 4 Anthropic · 32K effort | 34.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-4o mini | 34.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude 3.7 Sonnet Anthropic · 64K effort | 35.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Opus 4.1 | 35.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 nano OpenAI · Medium effort | 35.4% | Third party | VPCT leaderboard via Epoch AI | — |
| o1 OpenAI · Medium effort | 37.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 nano OpenAI · High effort | 37.2% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Opus 4 Anthropic · 16K effort | 38.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Sonnet 4.5 | 38.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Flash | 38.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude 3.7 Sonnet | 39.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 mini OpenAI · High effort | 39.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Sonnet 4.5 Anthropic · 32K effort | 39.8% | Third party | VPCT leaderboard via Epoch AI | — |
| Claude Opus 4.5 Anthropic · 32K effort | 40.0% | Third party | Tweet from VPCT creator via Epoch AI | — |
| GPT-4o | 40.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 mini OpenAI · Medium effort | 40.2% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Pro | 40.5% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini Robotics-ER 1.5 | 40.8% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-4.5 | 45.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Flash | 46.2% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Pro | 46.4% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 2.5 Pro | 48.0% | Third party | VPCT leaderboard via Epoch AI | — |
| o3 OpenAI · Medium effort | 52.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5.1 OpenAI · Medium effort | 53.3% | Third party | VPCT leaderboard via Epoch AI | — |
| o4-mini OpenAI · Medium effort | 57.5% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5.1 OpenAI · High effort | 58.7% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 OpenAI · Medium effort | 63.2% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5 OpenAI · High effort | 66.0% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5.2 OpenAI · High effort | 67.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 3 Flash | 72.6% | Third party | VPCT leaderboard via Epoch AI | — |
| GPT-5.2 OpenAI · Xhigh effort | 84.0% | Third party | VPCT leaderboard via Epoch AI | — |
| Gemini 3 Pro | 91.0% | Third party | VPCT leaderboard via Epoch AI | — |
- Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0.