DownloadBook a demo

Layout defect scenario · 2026-08-13

Can vision models recognize the clean IoT control center UI?

Determine whether this clean control contains a visible layout defect.

Expected answerClean

1440 × 900 · 9 model configurations

Research by Stephan PetzlPublished

Exact input

Prompt

You are evaluating a user interface screenshot for objective rendering defects.

Check only for these visible issue categories: viewport_cropping, container_clipping, text_overflow, element_overlap, obscured_actionable_control, or broken_responsive_reflow. Do not report subjective design preferences, copywriting concerns, or accessibility assumptions.

If there is one visible issue, return its category and the tight affected region on a normalized 0 to 1000 grid. If none of these issues is visible, return has_issue=false and null category and region values.

Return only data matching the supplied JSON schema. Do not describe your reasoning.
SHA-256 · 7d8cfe281f4ef9b625bbcc6c89c45236b392624e96c23044e6dcaac1d8d7adb0

Ground truth

IoT control center layout-defect benchmark screenshot
Ground truth: clean control

Model evidence

9 model results

Blue marks the seeded defect region. Orange marks the model’s returned region when available.

Holo3 122B A10B

H Company · holo3-122b-a10b
Correct
Latency
1.48 s
Category
—
Localization IoU
—
Cost
$0.00063
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description": "No issues found."}

Grok 4.5

xAI · x-ai/grok-4.5
Correct
Latency
7.77 s
Category
—
Localization IoU
—
Cost
$0.0038
Attempts
1
                  {"has_issue": false, "category": null, "region": null, "description": "No objective layout defects detected."}

GPT-5.4

OpenAI · openai/gpt-5.4
Correct
Latency
9.04 s
Category
—
Localization IoU
—
Cost
$0.0058
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":"No visible rendering defect from the allowed categories."}

Gemini 3.6 Flash

Google AI Studio · google/gemini-3.6-flash
Incorrect
Latency
8.94 s
Category
—
Localization IoU
—
Cost
$0.0047
Attempts
1
                  {"has_issue

SyntaxError

GPT-5.6 Sol

OpenAI · openai/gpt-5.6-sol
Correct
Latency
7.26 s
Category
—
Localization IoU
—
Cost
$0.0128
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":"No visible rendering defect."}

GPT-5.6 Luna

OpenAI · openai/gpt-5.6-luna
Correct
Latency
8.01 s
Category
—
Localization IoU
—
Cost
$0.00028
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":""}

Claude Sonnet 5

Anthropic · anthropic/claude-sonnet-5
Correct
Latency
9.57 s
Category
—
Localization IoU
—
Cost
$0.0059
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":"No visible layout defects detected; all elements appear properly rendered without cropping, clipping, overflow, overlap, or reflow issues."}

Claude Opus 5

Anthropic · anthropic/claude-opus-5
Correct
Latency
9.99 s
Category
—
Localization IoU
—
Cost
$0.0140
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":"No visible rendering defects."}

Gemini 3.1 Pro Preview

Google AI Studio · google/gemini-3.1-pro-preview
Correct
Latency
22.27 s
Category
—
Localization IoU
—
Cost
$0.0027
Attempts
1
                  {"has_issue":false,"category":null,"region":null,"description":"No visible layout defects detected."}
EN