COHERENCE - Science: leaderboard

COHERENCE (2026) removes the images from 6,161 interleaved image-text documents (WikiHow, StoryBird, cooking recipes, science) and asks the model to put each candidate image back in its place. This board is the exact-match accuracy on the science documents, in percent; higher is better.

Metric: Exact-match accuracy (%) on the 1,361 science documents; 6,161 interleaved image-text documents whose images were removed: the model assigns each candidate image to its placeholder; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.4High68.41#135
2Gemini 3.1 Pro (Preview)65.1#99
3Qwen 3.5 397B A17B60.32#264
4Claude Sonnet 4.6Thinking58.93#173
5Kimi K2.552.09#266
6Doubao-Seed-2.0-Pro-26021549.01#158
7Qwen 3.5 122B A10B48.79#346
8Doubao-Seed-2.0-Lite-26021545.78#262
9Qwen 3.5 35B A3B44.09#492
10GPT-5.240.78#185
11Doubao-Seed-2.0-Mini-26021537.62
12Intern-S1-Pro35.19#585
13Qwen 3 VL 30B A3B34.39#784
14Qwen 3 VL 235B A22B32.77#485
15Qwen 3.5 4B31.45#1241

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

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