Back to feed
arXiv cs.AI·

Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

Signal
78
Hype
25
In three linesStudy measuring no-CoT reasoning capability across 30,000+ questions spanning 43 benchmarks. Frontier models double their 50%-task-completion time horizon yearly: GPT-5.5 reaches 3+ minutes without explicit reasoning tokens. Projections: 7 minutes by 2028, 25 minutes by 2030.
Read source
Your take?
ReasoningBenchmarksAI safetyGPT

Summary generated by Claude — human-verified