ARC-AGI-3 leaderboard: models score 90%+ but at $10K+ per task
haider1 · x · 2026-11-12
Haider shares the verified ARC-AGI-3 leaderboard, commenting that the LLM paradigm may be hitting a wall — and this time he's more worried about the wall. The chart shows Claude Fable 5.2, GPT-6 Astra and Gemini 4 Argon reaching 60–95% scores, but top configurations cost thousands to tens of thousands of dollars per task, with most zero-score attempts still clustered at hundreds to thousands of dollars — a brutal performance-to-cost ratio.
More from Models
- Polymarket odds put Chinese AI lab at #1 by year-end at just 6% — Polymarket · 2026-11-12
- FutureHouse's deslopper trains Kosmos to write indistinguishably from human experts — proteinrosh · 2026-11-12
- Developers say Gemini 4 Carbon is notably better at favicon design and micro-interactions — AI_Andrew · 2026-11-12
- Gemini Decision 1 tops knowledge and reasoning evals on calibration, at a steep cost — ArtificialAnlys · 2026-11-12
- Gemini Decision 1 costs $345 per 1M tasks, 3x pricier than leader Drex Max v1.1 — ArtificialAnlys · 2026-11-12
- Google's Gemini Decision 1 scores 66 on AA Decision Index, trails leaders at 3x the cost — ArtificialAnlys · 2026-11-12