Continual-ARC benchmark: leading agent frameworks fail to learn from experience
hey_abusiddik · x · 2026-11-12
Unify.ai's new Continual-ARC benchmark tests whether agents can learn from experience and generalize. Results: OpenClaw, Hermes Agent and Prime Agent memorized nothing of value and failed to generalize to new tasks.
The team's key point: memory that never reaches the model's context does nothing — most agent memory problems start there. As a contrast, fine-tuning a small 9B model's weights achieved both memorization and generalization, suggesting the current external-memory-plus-prompts paradigm has fundamental limits on continual learning.
More from coding & agent
- United Airlines to Ban Personal AI Agents From Booking Flights for Customers — Polymarket · 2026-11-12
- Agentic Search explained: a 4-search loop costs about 5x one search — and queries leak to more companies — adnan_hashmi · 2026-11-12
- AWS says thousands of autonomous AI agents wrote every line of AWS Emulator — SumitGup · 2026-11-12
- freeCodeCamp launches free 2-hour course on agentic coding with Codex — adnan_hashmi · 2026-11-12
- A criminally underrated resource comparing tool-calling reliability across models and providers — ricklamers · 2026-11-12
- Organizing 55k data points is easy: enterprise data prep for agents is the hard part — BenBajarin · 2026-11-12