Benchmarks · /benchmarks/scale-vtb

VTB

Scale VTB leaderboard for image-grounded reasoning and multimodal understanding.

Source · Scale Labs
Version · scale-labs snapshot 2026-05-01
Scores · 12

Passport

Visible tradeoffsThis is a rubric-judged signal, so it is more structured than arena taste but still depends on the scoring rubric.

source

Scale Labs

metric

APR (%)

judge

Rubric

direction

higher better

group id

scale_vtb_current

domain

Vision understanding

What it measures vs what it misses

✓ Measures

Visual interpretation and reasoning over benchmark images and prompts.

✗ Misses

Image generation quality. Tool-use orchestration beyond the judged task.

Why this countsIt is useful when the model must read charts, UI, screenshots, or visual scenes rather than text alone.Comparable-group ruleThis percentile only compares models inside the exact benchmark/version group shown here. It is not a universal score.What it missesIt does not tell you whether the model can generate or edit images well.

Leaderboard · this benchmark version

#1 · GPT-5.4

SL · Apr 29, 2026

29.2%

#2 · Gemini 3.1 Pro

SL · Apr 29, 2026

29%

#3 · Claude Opus 4.6

SL · Apr 29, 2026

27.5%

#4 · Gemini 3 Pro Preview

SL · Apr 29, 2026

26.9%

#5 · GPT-5

SL · Apr 29, 2026

17%

#6 · GPT-5.4 mini

SL · Apr 29, 2026

17%

#7 · GPT-5.4 nano

SL · Apr 29, 2026

17%

#8 · o3

SL · Apr 29, 2026

13.7%

#9 · Gemini 2.5 Pro

SL · Apr 29, 2026

11.8%

#10 · Claude Sonnet 4

SL · Apr 29, 2026

4.5%

#11 · Claude Sonnet 4.5

SL · Apr 29, 2026

4.5%

#12 · Claude Sonnet 4.6

SL · Apr 29, 2026

4.5%