AgentDrive - Physics: leaderboard
Evaluates text reasoning about autonomous-driving scenarios on the 2,000-example AgentDrive-MCQ evaluation. Published model/version labels are preserved. This series reports Physics using Accuracy (%). Source: the original paper result table.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Not forecast. 51 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B 2507 | 67.5 | #341 |
| 2 | Claude 3.7 Sonnet (Thinking) | 65 | #259 (Claude 3.7 Sonnet) |
| 3 | Llama 3.1 70B Instruct | 57.5 | #567 |
| 4 | Hermes-3-Llama-3.1-405B | 57.5 | #531 |
| 5 | Llama 3.3 70B Instruct | 55 | #541 |
| 6 | DeepSeek V3.2 Exp | 55 | #241 |
| 7 | ChatGPT 4o | 55 | #296 |
| 8 | GPT-4.1 | 52.5 | #258 |
| 9 | GPT-4o Mini | 52.5 | #598 |
| 10 | GPT-4o Mini (2024-07-18) | 52.5 | #568 |
| 11 | Mistral Medium 3.1 | 52.5 | #418 |
| 12 | ERNIE 4.5 300B A47B | 52.5 | #478 |
| 13 | GPT-4.1 Mini | 50 | #363 |
| 14 | Gemma 3 12B (IT) | 50 | #672 |
| 15 | GPT-5 Chat | 50 | #250 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=agentdrive-physics · How It Works · Data refreshed daily, snapshot 2026-11-05.