SciAgentBench - Physics: leaderboard
Metric: Success rate (%, 0-100) on the 109 physics tasks; with tools: a ReAct loop over the task-relevant SciAgentGym tools (native function calling, up to 50 tool rounds, 300-second request timeout); SciAgentBench: 259 multi-step scientific tasks (1,134 sub-questions; physics 109, chemistry 81, materials 37, life sciences 32; about 65 percent with images) aggregated from existing benchmarks, kept when four frontier LLMs averaged under 50 percent and SciAgentGym could execute a verified trace; a task counts only when every sub-question is correct (strict JSON matching with 0.05 numeric tolerance, GPT-4.1 checking textual fields); temperature 0.7; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 | 47.2 | #218 |
| 2 | GPT-5 | 46.3 | #91 |
| 3 | Claude Sonnet 4 | 39.4 | #194 |
| 4 | Gemini 2.5 Flash | 38.3 | #237 |
| 5 | Gemini 2.5 Pro | 37.3 | #145 |
| 6 | O3 | 35.5 | #121 |
| 7 | Qwen 3 VL 32B (Thinking) | 33 | #287 (Qwen 3 VL 32B) |
| 8 | Qwen 3 VL 32B Instruct | 31.8 | #276 |
| 9 | O4 Mini | 31.2 | #172 |
| 10 | GLM-4.6V | 30.9 | #309 |
| 11 | Qwen 3 VL 235B A22B (Thinking) | 30.6 | #228 (Qwen 3 VL 235B A22B) |
| 12 | Qwen 3 VL 235B A22B Instruct | 28.1 | #264 |
| 13 | Qwen 3 VL 8B Instruct | 24 | #401 |
| 14 | Qwen 3 VL 4B Instruct | 23.8 | #506 |
| 15 | GPT-4o | 21.3 | #333 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=sciagentbench-physics · How It Works · Data refreshed daily, snapshot 2026-10-11.