ARC-AGI cost collapse: DeepSeek-V4-Flash hits 87% at $0.021/task vs o3's $4,500
inductionheads · x · 2026-09-12
A viral comparison of ARC-AGI-1: o3-preview scored 87.5% in Dec 2024 at over $4,500 per task; DeepSeek-V4-Flash scored 87.0% in July 2026 at just $0.021 per task — a 200,000x cost drop in 18 months.
The author argues OpenAI merely pulled forward open-source capability by a year or two, and that by 2028 open flash models will crack such problems in under an hour, with no way to police it — mathematics as a profession must change. Context: 25 Fields Medal winners issued a joint declaration warning of severe misalignment between AI companies and the math community.
More from AGI Musings
- Google's recursive self-improvement push: training kernel 23% faster, TPU gains — imjustnewatai · 2026-09-12
- Can machines prove everything? Turing, Gödel and the practical limits of AI mathematics — jamestagg · 2026-09-12
- Biggest AI risk PR funder began as global health funder, argues KOL — kevinnbass · 2026-09-12
- Tucker Carlson records podcast with AI risk researcher Nate Soares on AI 'killing everyone' — Polymarket · 2026-09-12
- 'Staff' once meant your own walking stick — AI agents should be loyal to you, not your company — granawkins · 2026-09-12
- The AI paradox: insiders building frontier models want to slow down, outsiders don't — CFGeek · 2026-09-12