VPCT
Visual physics comprehension: predict where the ball lands.
Agent Index weight
0%
shown for context only
Models tested
24
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Gemini 3 Pro Preview Supported Google | 91.0% | source | — | — | |
| 2 | GPT-5.2 Supported OpenAI | 84.0% | source | — | — | |
| 3 | Gemini 3 Flash Preview Supported Google | 72.6% | source | — | — | |
| 4 | GPT-5 Supported OpenAI | 66.0% | source | — | — | |
| 5 | GPT-5.1 Supported OpenAI | 58.7% | source | — | — | |
| 6 | o4-mini Estimated OpenAI | 57.5% | source | — | — | |
| 7 | o3 Supported OpenAI | 52.0% | source | — | — | |
| 8 | Gemini 2.5 Pro Supported Google | 48.0% | source | — | — | |
| 9 | Gemini 2.5 Flash Estimated Google | 46.2% | source | — | — | |
| 10 | GPT-4.5 OpenAI | 45.0% | source | — | — | |
| 11 | gemini-robotics-er-1.5-preview Unknown | 40.8% | source | — | — | |
| 12 | GPT-5 Mini Estimated OpenAI | 40.2% | source | — | — | |
| 13 | Claude Opus 4.5 (latest) Supported Anthropic | 40.0% | source | — | — | |
| 14 | GPT-4o (2024-08-06) Supported OpenAI | 40.0% | source | — | — | |
| 15 | Claude Sonnet 4.5 (latest) Supported Anthropic | 39.8% | source | — | — | |
| 16 | Claude 3.7 Sonnet Supported Anthropic | 39.0% | source | — | — | |
| 17 | Claude-Opus-4 Supported Anthropic | 38.0% | source | — | — | |
| 18 | GPT-5 Nano OpenAI | 37.2% | source | — | — | |
| 19 | o1 Estimated OpenAI | 37.0% | source | — | — | |
| 20 | Claude Opus 4.1 (latest) Supported Anthropic | 35.0% | source | — | — | |
| 21 | Claude Sonnet 4 Supported Anthropic | 34.0% | source | — | — | |
| 22 | GPT-4o mini OpenAI | 34.0% | source | — | — | |
| 23 | Claude-Sonnet-3.5 Estimated Anthropic | 33.0% | source | — | — | |
| 24 | Gemini 2.5 Flash-Lite Google | 30.0% | source | — | — |