PRO-LONG boosts coding agents 18 points on ARC-AGI-3 with lossless memory
burny_tech · x · 2026-07-29
PRO-LONG proposes programmatic memory for long-horizon agents by keeping a full structured interaction log and letting the agent search it with tools like grep, regex, and Python. On ARC-AGI-3, it improves base coding agents by 18 points on average and uses 4.2–5.8× fewer tokens than specialized harnesses.
The paper argues that summary-based memory loses details that may matter later. Instead, it keeps the trajectory lossless and searchable, which makes long-context reasoning more reliable and efficient.
Related event: PRO-LONG Memory Framework Achieves High ARC-AGI-3 Score at Low Cost(3 posts)→
More from coding & agent
- theo builds his own visualizer for today's agent models, showing how cheap Luna really is — ivan_bezdomny · 2026-09-23
- Vite+ Hits RC: One Rust-Powered CLI to Replace Your Entire Web Toolchain — cnakazawa · 2026-09-23
- Tesla's in-car Grok agent books trips across Gmail, Calendar and Notion in one command — xiaohu · 2026-09-23
- Tesla's In-Car Grok Assistant Now Executes Cross-App Tasks in One Sentence — xiaohu · 2026-09-23
- Garry Tan says Capy lets him ship PRs much faster than Codex or Claude Code — garrytan · 2026-09-23
- DeskPilot: open-source native Python desktop client for local LLMs with MCP and sandboxed tools — poofph · 2026-09-23