ReactBench (Reaction Diagrams) - Reasoning: leaderboard

ReactBench chemical reaction diagram questions that classify the overall topology (linear, branched, tree, cyclic graph) and detect cycles or terminal nodes. Exact-match accuracy in percent, higher is better. It is unrelated to the ReactBench React UI coding benchmark.

Metric: Answer accuracy (%) on the structural topology reasoning questions (classify the global topology, detect cycles and terminal nodes), 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 July 2027. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)64.85#99
2Qwen 3 VL 32B Instruct57.43#606
3Gemini 1.5 Pro55.45#1063
4Qwen 2.5 VL 72B54.46#781
5GPT-5.551.49#54
6GPT-4o49.5#751
7Claude 3.5 Sonnet49.5#752
8Claude Sonnet 4.646.53#173
9Qwen 3 VL 32B (Thinking)43.07#573
10MiniCPM-o-2.639.6#2423
11Qwen 2.5 VL 7B39.11#1862
12Phi-3.5-Vision Instruct38.12#6268
13InternVL2.5-8B-MPO37.62#2012
14Qwen 2.5 VL 3B34.65#2639
15DeepSeek-VL2-Tiny33.66#6534

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=reactbench-reaction-diagrams-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.