COHERENCE - Partial Match: 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 partial-match score, the Kendall tau between the predicted and true order rescaled to 0-100; higher is better.

Metric: Partial-match score (Kendall's tau between predicted and true order, rescaled to 0-100), over all 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
1Gemini 3.1 Pro (Preview)90.11#99
2Claude Sonnet 4.6Thinking88.76#173
3Qwen 3.5 397B A17B88.37#264
4Kimi K2.586.74#266
5GPT-5.4High86.54#135
6Qwen 3.5 122B A10B86.52#346
7Qwen 3.5 35B A3B85.06#492
8Doubao-Seed-2.0-Lite-26021584.79#262
9Doubao-Seed-2.0-Pro-26021584.76#158
10Doubao-Seed-2.0-Mini-26021583.27
11Qwen 3 VL 235B A22B81.61#485
12Qwen 3.5 4B81.16#1241
13Intern-S1-Pro80.56#585
14Qwen 3 VL 30B A3B79.5#784
15GLM-4.6V79.29#710

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

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