PRO-LONG Boosts Long-Horizon Agents with Programmable Memory

The PRO-LONG framework enhances long-horizon LLM agents by storing full interaction logs as searchable, structured "programmable memory" instead of compressing them. This approach significantly improves agent stability, achieving an 18-point gain on the ARC-AGI-3 benchmark.

2026-07-24 ~ 2026-07-24 · 2 related posts

• PRO-LONG gives LLM agents searchable programmatic memory and lifts ARC-AGI-3 by 18 points — dair_ai · 2026-07-24
• PRO-LONG keeps full action logs and lifts long-horizon agents by 18 points on ARC-AGI-3 — rohanpaul_ai · 2026-07-24