Benchmark RadarAI BENCHMARK PROFILE
GEB-Bench
GEB-Bench evaluates models on abstract structural motifs (e.g., self-reference, strange loops) presented in multiple modalities (natural scenes, folk stories, math theorems, programmatic skeletons) with tasks probing cross-modal mapping.
- Released
- 2026-08-04
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark aims to measure abstraction and cross-modal transfer, which are foundational for general intelligence but not addressed by current benchmarks.
Motivation
Can a model look at a river delta and a lightning bolt and see that they share a structure?
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.