ReactBench (Reaction Diagrams) - Tracing: leaderboard

ReactBench chemical reaction diagram questions that trace connectivity: the next node on a pathway or a complete path through the reaction network. Exact-match accuracy in percent, higher is better. It is unrelated to the ReactBench React UI coding benchmark.

Metric: Answer accuracy (%) on the pathway connectivity tracing questions (next nodes and complete paths through the reaction network), 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: Estimated already saturated. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude 3.5 Sonnet92.22#752
2Qwen 3 VL 32B Instruct91.02#606
3Claude Sonnet 4.690.42#173
4Qwen-VL-Max (ReactBench checkpoint unspecified)88.62
5Qwen 3 VL 32B (Thinking)88.02#573
6Gemini 3.1 Pro (Preview)86.23#99
7Qwen 2.5 VL 72B83.83#781
8GPT-5.583.23#54
9GPT-4o82.04#751
10Qwen 3 VL 8B Instruct82.04#1029
11Qwen 3 VL 8B (Thinking)76.65#910
12Gemini 1.5 Pro71.86#1063
13InternVL2.5-78B67.66#1031
14Qwen 2.5 VL 7B66.47#1862
15InternVL2.5-26B65.87#2156

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-tracing · How It Works · Data refreshed daily, snapshot 2026-11-21.