Blueprint-Bench 2
50 apartments, each shown as about 20 interior photos. The agent draws a 2D floor plan of the rooms, how they connect and their relative sizes, which is compared with the real layout. The published score is normalized so that a random plan reads 0 and a perfect one 100%.
Understanding photos, diagrams and documents alongside text. One of 4 comparable tests in multimodal, a category the Index averages.
Read with care. Scores at or below the random baseline are published as zero. Andon Labs measured its human baseline, about 59%, on 12 of the 50 apartments only.
Publisher: Andon LabsWhat a model is asked to do
Turn about twenty interior photos of an apartment into a floor plan: which rooms there are, how they connect and how big they are.
For exampleA red armchair appears in two photos taken from different doorways. Use it to work out that the hallway links the living room and the study, then draw the plan.
Why it matters. Building a map of a place from pictures is basic spatial sense, the kind a model needs before it can help plan a renovation or guide a robot.
The example is original and illustrative, not an item from the dataset.
- Models scored here
- 36
- Who produced the numbers
- Official leaderboard
- Items graded
- n = 50 (2.0% each)
- Best published result
- 58.6%
- Newest release covered
- October 28, 2026 (active)
- License
- CC BY 4.0 (via Epoch AI)
- Direction
- Higher is better
- Scale
- Published as a share of items solved. For the Index, each score is a share of the best published result (58.6%), so the frontier reads 100. Guessing earns nothing on this test, so no chance correction applies.
- Provenance
- Produced by the benchmark's own leaderboard or a third-party evaluator, not by the model's maker.
Snapshot November 10, 2026
Ranking on this test
36 models on Blueprint-Bench 2
Best published run per model, and how close each one comes to the best published result on this test (the frontier reads 100). Every number links to the publisher that produced it. Internal runs are marked and carry their run details on the model page. One item on this test is 2.0 points, so read gaps smaller than that as noise.
The best models, ranked here, working inside Theo.
29 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.