Benchmark RadarEGOSTREAM
Egostream evaluates streaming episodic memory in egocentric vision through 2,250 curated questions across seven cognitive dimensions, expanded into 8,528 recall-conditioned evaluations using the Answer Validity Window to test recall from instant to ultra-long-term.
- Released
- 2026-05-29
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
Existing streaming video benchmarks lack fine-grained diagnosis of memory retention over time. Egostream provides a controlled protocol that separates genuine forgetting from world-state changes, enabling precise assessment of model memory capabilities across different temporal scales.
Motivation
Continuous episodic memory is a core capability for autonomous agents operating in dynamic, real-world environments, yet current streaming video benchmarks provide limited tools for diagnosing what models remember and for how long.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.