ReactBench (Reaction Diagrams): leaderboard
Metric: Answer accuracy (%) on the four task dimensions (unweighted mean of localization, extraction, tracing and reasoning accuracy), over the 1,618 expert-annotated question-answer pairs on more than 1,300 real chemical reaction diagrams from chemistry journals and patents of ReactBench (single-line, multi-line, tree and graph topologies), multiple-choice or short open answers matched exactly against the key after template extraction, no LLM judge; higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 32B Instruct | 73.91 |
| 2 | Gemini 3.1 Pro (Preview) | 73.88 |
| 3 | Qwen 3 VL 32B (Thinking) | 70.89 |
| 4 | GPT-5.5 | 69.59 |
| 5 | Claude 3.5 Sonnet | 68.6 |
| 6 | Claude Sonnet 4.6 | 67 |
| 7 | GPT-4o | 63.83 |
| 8 | Qwen 3 VL 8B Instruct | 63.02 |
| 9 | Gemini 1.5 Pro | 62.17 |
| 10 | Qwen 3 VL 8B (Thinking) | 60.24 |
| 11 | Qwen 2.5 VL 7B | 57.32 |
| 12 | InternVL2.5-78B | 53.74 |
Interactive version: theaggregate.ai/benchmark?slug=reactbench-reaction-diagrams · How It Works · Data refreshed daily, snapshot 2026-10-07.