CharXivBrowse 296

CharXiv

Can the model extract and reason over information in images or documents?

Version not specifiedExact reported variant
Vision14 ranked models17 reported values3 reportsHigher is better

Top rankings

One row per model, using its best reported score across effort settings.

14
RankModelBest scoreBest reported setting
1Claude Fable 5Anthropic88.9%max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source
2Kimi K3Moonshot AI84.8%max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source
3GPT-5.6 SolOpenAI84.6%max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source
4GPT-5.5OpenAI84.1%xhigh effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source
5Qwen3.5-397B-A17BQwen80.8%Reported configurationQwen report1 value · 1 reportApr 21, 2026 · source
6Claude Opus 4.8Anthropic80.5%max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source
7Qwen3.5-27BQwen79.5%Reported configurationQwen report2 values · 2 reportsApr 21, 2026 · source
8Qwen3.6-27BQwen78.4%Reported configurationQwen report1 value · 1 reportApr 21, 2026 · source
9Qwen3.6-35B-A3BQwen78.0%Reported configurationQwen report2 values · 2 reportsApr 21, 2026 · source
10Qwen3.5-35B-A3BQwen77.5%Reported configurationQwen report1 value · 1 reportApr 15, 2026 · source
11Gemma4-26B-A4BGoogle69.0%Reported configurationQwen report1 value · 1 reportApr 15, 2026 · source
12Claude Opus 4.5Anthropic68.5%Reported configurationQwen report1 value · 1 reportApr 21, 2026 · source

Effort curve

Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.

0
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology

Can the model extract and reason over information in images or documents?

Visual perception and multimodal reasoning under the reported protocol. Higher is better. The value is the percentage reported in this lab's table.

Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.

Method / source