ReactBench (Reaction Diagrams) - Extraction: leaderboard

ReactBench chemical reaction diagram questions that extract reaction conditions, yields and other attributes while keeping them tied to the right graph element. Exact-match accuracy in percent, higher is better. It is unrelated to the ReactBench React UI coding benchmark.

Metric: Answer accuracy (%) on the topological information extraction questions (reaction conditions, yields and labels tied to graph elements), 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 February 2027. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen-VL-Max (ReactBench checkpoint unspecified)90.1
2Qwen 2.5 VL 72B89.86#782
3Qwen 2.5 VL 3B89.61#2664
4Qwen 3 VL 32B (Thinking)88.89#577
5GPT-5.588.41#56
6Qwen 3 VL 32B Instruct86.96#607
7Qwen 3 VL 8B (Thinking)85.99#924
8Qwen 2.5 VL 7B85.75#1878
9Gemini 3.1 Pro (Preview)85.27#102
10Qwen 3 VL 8B Instruct85.02#1043
11Claude 3.5 Sonnet81.88#755
12GPT-4o79.71#754
13InternVL2.5-78B79.47#1036
14DeepSeek-VL2-Small78.5#2523
15Gemini 1.5 Pro75.6#1071

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=reactbench-reaction-diagrams-extraction · How It Works · Data refreshed daily, snapshot 2026-11-22.