Databricks Says MemCurrent Lifts Agent Pass Rate to 68.4% from 22.3%
Summary
Databricks AI Research Team introduces MemCurrent, a persistent memory architecture now running inside Agent Bricks Memory and Genie. The company says it lifts agent pass rates on a 114-case internal benchmark from 22.3% to 68.4% versus no memory, while cutting token use 1.44x and running 1.35x faster.
Key Points
- Databricks reports MemCurrent averages 88.1% across LoCoMo, LongMemEval and ConvoMem, matching full context at roughly 12x fewer input tokens.
- The 114-case internal benchmark draws on more than 12,000 Genie sessions from 29 users, with 2,233 history sessions.
- Databricks says offline dreaming alone reaches a 63.5% pass rate, versus 53.9% for online memory tools.