ComplexityWorld: leaderboard
ComplexityWorld (2026) renders hidden structured decision problems as maps, diagrams, boards and grids across 39 visual worlds, and a vision-language model must return a complete decision in one direct response. The verifier acceptance rate is the share of the 390 main-panel tasks whose decision an executable verifier accepts, in percent; higher is better.
Metric: Verifier acceptance rate (%; share of the 390 frozen main-panel tasks, 10 in each of 39 domain-inspired visual worlds and 29 decision categories, whose complete decision an executable verifier accepts as feasible; direct inference, one response from the image, instruction and output format, no agent loop or tools). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (xHigh) | 75.6 | #29 (GPT-5.6 Sol) |
| 2 | Qwen 3.7 Plus | 39.5 | #203 |
| 3 | Gemini 3.5 Flash | 32.3 | #98 |
| 4 | MiMo-V2.5 | 24.4 | #353 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=complexityworld · How It Works · Data refreshed daily, snapshot 2026-11-12.