VlmsAreBlind

VlmsAreBlind - benchmark testing visual perception capabilities that reveal blind spots in VLMs.

Category

Vision & Language

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for VlmsAreBlind

Model Score Date Verified
Qwen3.6-27B
100.00%
22.04.2026 Verified
Gemma4-31B-it
53.77%
22.04.2026 Unverified
Qwen3.6-35B-A3B
98.11%
22.04.2026 Unverified
Qwen3.5-27B
99.53%
22.04.2026 Unverified
Qwen3.5-4B
79.25%
01.02.2026 Verified
Qwen3.5-122B-A10B
98.58%
01.02.2026 Verified
Qwen3.5-35B-A3B
100.00%
01.02.2026 Unverified
Claude-Sonnet-4.5
45.75%
01.02.2026 Unverified
GPT-5-mini 2025-08-07
75.80
01.02.2026 Unverified
Qwen3-235B-A22B
17.45%
01.02.2026 Unverified
Qwen3-VL-235B-A22B
17.45%
01.02.2026 Verified