Agent Memory Challenge 2026 Cycle 2 Opens with Three Evaluation Tracks

Summary

Agent Memory Challenge 2026 Cycle 2 is now open for evaluations of long-term memory in AI agents. The organizers frame memory as more than storing additional history: an agent must retrieve the right evidence, recognize what has changed, and avoid relying on stale context when taking action. The challenge has three tracks, covering textual, coding, and multimodal settings. It accepts both open-source methods and commercial products. Eligible open-source teams can compete for a prize pool exceeding $22,000. Participants use a shared Add/Search interface and a standardized Answer/Eval process, with public results intended to support direct comparison. The announcement invites teams to join through agentmemoryleaderboard.ai/evaluation.