MortgageTax: leaderboard

Evaluating reading and understanding tax certificates as images.

Metric: Accuracy (%). Source: www.vals.ai. Status: years away from saturation. 98 models tracked.

Top models

#ModelScore
1Claude Opus 572.06
2Claude Fable 5.1 (Max)70.79
3Claude Opus 4.7 (Max)70.27
4Claude Sonnet 5 (Max)70.03
5Claude Opus 4.8 (Max)69.91
6Gemini 3.1 Pro (Preview) (High)69.4
7Gemini 3 Pro (Preview) (High)69.08
8Gemini 2.5 Pro68.92
9Claude Fable 5 (Max)68.92
10GPT-5.5 (xHigh)68.76
11Gemini 3 Flash (Preview) (High)68.72
12Claude 3.7 Sonnet (20250219)68.68
13Gemini 3.5 Flash Lite (High)68.68
14Claude Opus 4.5 (20251101) (High)68.68
15Claude Opus 4.6 (Adaptive Reasoning, Max Effort)68.52

Interactive version: theaggregate.ai/benchmark?slug=mortgagetax · How It Works · Data refreshed daily, snapshot 2026-09-05.