Google Research·· 2026-04-22
ReasoningBank: Enabling agents to learn from experience
ReasoningBank: Enabling agents to learn from experience
AI summary
Google Research released the ICLR paper ReasoningBank, an open-source agent memory framework that distils high-level reasoning patterns from successful and failed experiences.
Selection record
Threshold 60Official, first-handFirst 62Second 62
AdmittedSum of both 124 ≥ twice the threshold 120
- Source tier
- Official, first-hand; this tier's threshold is 60
- Pre-filter
- passed：Google Research发布Agent记忆框架ReasoningBank
- Why it was chosen
- Success rates and step-count changes on WebArena and SWE-Bench-Verified support comparisons with existing memory approaches.
A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.
Source: Google Research · research.google