Observed arrival · 2026-11-08
Continual-ARC makes AI learning a recurring exam
A lifelong-learning benchmark built from ARC-AGI tasks, where AI learners revisit tasks, spend feedback budgets, and face post-stream tests.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Continual-learning researchers and AI benchmark builders
- Worth noticing
- Learners are frozen after the stream, then tested on fresh instances of both encountered and unseen tasks.
Field notes
The benchmark cycles through tasks from three ARC-AGI editions, returning to familiar tasks with fresh instances while introducing new ones. Learners can use demonstrations, answer attempts, or a harness, but feedback has a budget cost; evaluation then separates retention from generalization. The page reports 114 runs and compares harness-based approaches with prompting, program induction, and weight-based learning.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
ƒJavaScriptBrowser-side code central
One card from the complete issue