SciAgentBench - L2 (4 to 7 Steps): leaderboard
Metric: Success rate (%, 0-100) on the tasks of difficulty L2 (4 to 7 steps); 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 | 43.9 | #218 |
| 2 | GPT-5 | 38.4 | #91 |
| 3 | Claude Sonnet 4 | 36.5 | #194 |
| 4 | O4 Mini | 35 | #172 |
| 5 | Gemini 2.5 Pro | 33.6 | #145 |
| 6 | Gemini 2.5 Flash | 33.1 | #237 |
| 7 | O3 | 31.4 | #121 |
| 8 | GLM-4.6V | 27.5 | #309 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 27.4 | #228 (Qwen 3 VL 235B A22B) |
| 10 | Qwen 3 VL 32B Instruct | 27.3 | #276 |
| 11 | Qwen 3 VL 32B (Thinking) | 27.3 | #287 (Qwen 3 VL 32B) |
| 12 | Qwen 3 VL 8B Instruct | 25.2 | #401 |
| 13 | Qwen 3 VL 235B A22B Instruct | 22.6 | #264 |
| 14 | GPT-4o | 17.4 | #333 |
| 15 | Qwen 3 VL 4B Instruct | 16.5 | #506 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=sciagentbench-l2-4-to-7-steps · How It Works · Data refreshed daily, snapshot 2026-10-11.