0

Blueprint-Bench 2

Frontier

Spatial-reasoning benchmark over architectural blueprints (version 2).

Domain
spatial-reasoning
Published
Jun 2026
Notable for
Spatial-reasoning benchmark used at Claude Fable 5 launch.

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
Anthropic
Attribution policy →

Top score 38.6% by Claude Fable 5 - 4 models reporting (3 frontier)

Score history

4
0%25%50%75%100%Feb 26Mar 26Apr 26May 26Jun 26Gemini 3.1 Pro PreviewGPT-5.5Claude Fable 5

Top models

4
Blueprint-Bench 2Bar chart with 4 bars. Highest value: Claude Fable 5 at 38.6.
4 models

FAQ

What is Blueprint-Bench 2?
Spatial-reasoning benchmark over architectural blueprints (version 2).
What is the current top score on Blueprint-Bench 2?
The top reported score is 38.6% by Claude Fable 5, across 4 models reporting (3 from frontier labs).