KnotBench - PD Code Choice: leaderboard

The D1 task of KnotBench (2026, The Gordian Knot for VLMs): judge which of four PD codes describes a rendered diagram, with answers checked against Regina canonical knot signatures. The score is accuracy in percent under strict answer matching, with a random baseline of 25 percent; higher is better.

Metric: Accuracy (%) on the 200 D1 items: which of four PD codes describes a rendered diagram; zero-shot with a forced ANSWER line and strict matching, 64K output-token budget, random baseline 25%; higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 2 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.7Thinking35.5#85
2GPT-5Thinking30.5#178

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

Interactive version: theaggregate.ai/benchmark?slug=knotbench-pd-code-choice · How It Works · Data refreshed daily, snapshot 2026-11-21.