Best for
- • Long-horizon memory-guided tool use
Memory benchmark
DolphinBench
An open benchmark for long-term agent memory that evaluates whether a system uses years of conversation history to take the correct tool action, while also reporting cost and latency.
Evaluation style
- • Tool-using agent tasks in simulated apps
Memory horizon
- • Nearly five years of conversation history
Scale
- • 600 verified tasks across three simulated users with roughly 500K tokens of message history per persona.
Tasks
- • Memory-guided tool use
- • Long-horizon rule application
- • Stateful application actions
What it measures
- • Task accuracy
- • Total cost
- • Median latency
Original research
DolphinBench: Mapping the Pareto Frontier of Agent Memory
This is the paper that introduced DolphinBench.
arXiv:2609.24971 ↗