Benchmark RadarAI BENCHMARK PROFILE
Generative Embedding Benchmark
GEB evaluates embeddings by measuring answer-relevant content recoverable by a decoder, using a visual question-answering dataset with development and test splits, and scoring answer quality.
- Released
- 2026-08-07
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Fills a gap by measuring generative information in embeddings, which is not captured by separability-based benchmarks, providing insight into information preservation for downstream generation.
Motivation
Embeddings have emerged as a standard representational interface linking foundation models with downstream systems.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.