Experiment · active

Counting and spatial relations

Can the model obey exact object counts and left-right-inside-behind relationships?

Runs
5
Variable
spatial constraint
Prompt
spatial-counting-prompt v1.0

Protocol

Active round-one controlled comparison using one frozen prompt across five hosted model endpoints or quality tiers.

Frozen prompt template

CONTROL PROMPT AB-R1-SPATIAL-001. A wide 16:9 museum-catalog photograph of a matte charcoal tabletop containing exactly seven objects and nothing else: three red wooden cubes in one straight row on the left; two transparent glass spheres in the center; one small aged-brass pyramid directly behind the two spheres; and one folded cream paper crane on the far right. Camera elevated thirty degrees, all seven objects fully visible and separated, physically realistic materials, neutral softbox lighting, dark seamless background. No labels, no text, no watermark, no duplicate objects, no containers.

Review rubric

  • Prompt Adherence
  • Counting
  • Spatial Reasoning

Notes

One unseeded generation per model. Exact visible deviations are preserved in each run's review notes.

Evidence

Recorded runs

Generated specimen r1-spatial-grok-imagine r1-spatial-grok-imagine
successgrok-imagine-image

R1 Spatial Grok Imagine

Vision adherence score 10.0/10. Exact count and relations: 3 red cubes left, 2 glass spheres center, 1 brass pyramid behind them, 1 cream crane far right; 7 total.

Size
1280×720
Seed
not exposed
Method
first party tool
View experiment →
Generated specimen r1-spatial-minimax-image-01 r1-spatial-minimax-image-01
successimage-01

R1 Spatial Minimax Image 01

Vision adherence score 3.0/10. Count failure: 5 red cubes, 3 glass spheres, 1 brass pyramid, and 1 cream crane; 10 total. Left-to-right category order is otherwise preserved.

Size
1280×720
Seed
not exposed
Method
official api
View experiment →
Generated specimen r1-spatial-openai-gpt-image-2-high r1-spatial-openai-gpt-image-2-high
successgpt-image-2-high

R1 Spatial Openai Gpt Image 2 High

Vision adherence score 10.0/10. Exact count and relations: 3 red cubes left, 2 glass spheres center, 1 brass pyramid behind them, 1 cream crane far right; 7 total.

Size
1672×941
Seed
not exposed
Method
official api
View experiment →
Generated specimen r1-spatial-openai-gpt-image-2-low r1-spatial-openai-gpt-image-2-low
successgpt-image-2-low

R1 Spatial Openai Gpt Image 2 Low

Vision adherence score 10.0/10. Exact count and relations: 3 red cubes left, 2 glass spheres center, 1 brass pyramid behind them, 1 cream crane far right; 7 total.

Size
1672×941
Seed
not exposed
Method
official api
View experiment →
Generated specimen r1-spatial-openai-gpt-image-2-medium r1-spatial-openai-gpt-image-2-medium
successgpt-image-2-medium

R1 Spatial Openai Gpt Image 2 Medium

Vision adherence score 10.0/10. Exact count and relations: 3 red cubes left, 2 glass spheres center, 1 brass pyramid behind them, 1 cream crane far right; 7 total.

Size
1672×941
Seed
not exposed
Method
official api
View experiment →