Hype watch, mostly facts
past.dev claims top memory score on DolphinBench
Reality check
A press release, not an official Mem0 run. The score comes from a Hermes Agent setup with GPT 5.6 Luna.
What the research found
past.dev self-reports a top DolphinBench score (Mem0's agent action-memory benchmark, 600 simulated tasks) using Hermes Agent with GPT 5.6 Luna; the result is not yet an official Mem0 run and this is a press release.
Claims and how well the evidence supports them
- past.dev reported the highest score yet on DolphinBench as of Oct 20, 2026 98%
- DolphinBench has 600 simulated tasks across chat, email, calendars, documents and code reviews 98%
- past.dev ran the test itself with Nous Research Hermes Agent and GPT 5.6 Luna 98%
- The score is self-reported; past.dev asked Mem0 to run it as an official configuration 97%
Evidence
- finance.yahoo.com https://finance.yahoo.com/technology/ai/articles/memory-ai-agents-gets-more-125900759.html
- past.dev https://past.dev/benchmarks/dolphinbench
Actionability rated low.