Letta benchmarks five AI agent harnesses on FrontierMemory test

Letta released FrontierMemory, a benchmark measuring whether AI agents learn from past experience rather than just recalling facts. Letta Dream led with 73.3%, followed by Prime Agent at 60%, OpenAI Agents SDK at 40%, Claude Managed Agents at 33.3% and Hermes Agent at 26.7%.

• Letta Dream scored 73.3%, Prime Agent 60%, OpenAI Agents SDK 40%
• Claude Managed Agents and Hermes Agent trailed at 33.3% and 26.7%
• GPT-6.1 Sol cost $0.77 per task versus $2.80 for Opus 5.5
• Version 1.0 has only 15 tasks with one attempt per task