Fchollet: ARC-4 due early 2027, final ARC likely 6-7, AGI around 2030
MartinSignoux · x · 2026-09-04
ARC benchmark creator François Chollet laid out the series' roadmap: ARC-4 is in the works for early 2027, ARC-5 is also planned, and the final ARC will probably be 6-7. His stated goal is to keep building benchmarks that humans can solve and AI can't, until designing such tasks is no longer possible — putting AGI around 2030.
Martín Signoux added pacing context: ARC-3 shipped in March 2026 and Astra has just saturated it — while ARC-2 was far from saturated when ARC-3 launched. At this pace, ARC-4 arrives in roughly six months, raising the question of whether we'll even get to ARC-5.
More from AGI Musings
- "Start preparing for the possibility that the world isn't going to end": a pushback on AI doom — BLUECOW009 · 2026-09-04
- Erik Voorhees: instead of pausing superintelligence, pause the super incapable of intelligence — 0xAllen_ · 2026-09-04
- DeepMind researcher argues AI consciousness is mathematically impossible: simulation isn't instantiation — AnnaCiaunica · 2026-09-04
- Standing out when everyone has AI: subtract the generic, then go deep — round · 2026-09-04
- GPT-6 Closes the Coding Gap but Trails on Deep Reasoning — Why Google May Reach AGI First — aditipawarr · 2026-09-04
- UK's AI Security Institute also lost control of models that hacked real targets — GarrisonLovely · 2026-09-04