PRO-LONG at ARC Prize Summit Doubles GPT-5.6+Codex Cross-Episodic Memory on BALROG
xennygrimmato_ · x · 2026-10-24
At the ARC Prize summit in Boston (speakers incl. Greg Kamradt and François Chollet), a talk presented PRO-LONG, tested on NetHack for cross-episodic memory. On BALROG, GPT-5.6+Codex with PRO-LONG reaches mean best@k of 24.5 (vs 13.5 baseline) at 5 episodes, and Kimi+Mini-SWE improves from 5.1 to 11.3 — persistent cross-episode logs substantially extend agent horizons.
Related event: ARC Prize Research Summit held at MIT with Chollet and others(2 posts)→
More from coding & agent
- Claude Code helped uncover 14 speaker-muted AirTags hidden across his family's cars and bags — luisdans · 2026-10-24
- Should an AI call agent keep running when one tool goes down? A degraded-mode proposal — SubtleInterval_4449 · 2026-10-24
- Claude Desktop blocks local models; users bypass it with a loopback proxy trick — pneuny · 2026-10-24
- OpenAI ships Automations in Agents API for scheduled agent tasks — OpenAIDevs · 2026-10-24
- Arca runs its financial services business on long-horizon agents, not with them — AI Engineer · 2026-10-24
- AGENTS.md explained: one root Markdown file to rule every AI coding tool — adnan_hashmi · 2026-10-24