Look closer.Reason further.

A visual puzzle benchmark for multimodal agents.
From what an image shows to what it really means.

Real puzzles. Hidden rules. One answer to discover.

Small images.
Big reasoning challenges.

A collection of visual riddles where symbols, language and everyday knowledge meet. Each puzzle asks an agent to discover the rule before finding the answer.

Visual puzzles
932
Source collections
11
Tasks in the pilot
39

Many perspectives. One collection.

LILACPKUFPANazo Timeand more

Your turn
to look closer.

Start with the image. Look for a connection.
Reveal a hint when you need a different perspective.

A001

Many scripts, one character

The image contains the clues. Can you work out the hidden rule?

Reveal a mechanism hint
Reveal the answer

Current annotation; review is ongoing.

17 selected examples from the working collection. Original puzzle text is preserved; English titles are editorial labels.

A first look
at the results.

39 puzzles. Four model–agent configurations.
A preliminary view of visual puzzle solving.

Download CSV

Preliminary results on the 39-task pilot
Model / agent Reasoning Solved Time / task

Pilot results, not scores on all 932 puzzles. Historical runs use different native harnesses, time budgets and prompt revisions, so this is an exploratory comparison. Timeouts count as unsolved.

How to read these results

No-hint accuracy is the number of correct final answers divided by 39. The GPT and Doubao runs used difficulty-based limits of 10, 20 or 30 minutes; DeepSeek used a uniform 20-minute limit.

Hint retests were run only on baseline failures: GPT-6 Sol solved 12 of 21 retested tasks, and DeepSeek Flash solved 17 of 34. Combined coverage is (baseline correct + retest correct) / 39. It is not accuracy from a full 39-task hinted run.

GPT-6.2 Sol and Doubao do not have hint-retest results in this snapshot. Doubao recorded 17 timeouts and DeepSeek recorded 4 in the no-hint pilot.

The question is visual.
The test is reasoning.

A controlled task package, a fresh agent session, and a final answer you can check.

1

See the puzzle.

Each task provides one image, a title when available, and permitted auxiliary text. Reference answers stay separate.

2

Find the mechanism.

Agents keep their native tools. Background research is permitted; searching for the puzzle’s existing answer is prohibited.

3

Commit to an answer.

Final answers are matched against accepted references. Failed baseline tasks may be retested with mechanism hints in fresh sessions.

Search restrictions are prompt-based and require trace review. They do not by themselves prove that an agent avoided existing solutions.

Take a closer look.

Explore the preview data and inspect the pilot results.

Research preview · Full dataset release details are not yet available.

A001

Enlarged puzzle