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

Researchers measured how well frontier AI models perform complex reasoning without explicit chain-of-thought (CoT) tokens, finding that no-CoT task-completion time horizons have doubled yearly over six years. GPT-5.5 now reaches over 3 minutes of reasoning complexity, with projections suggesting frontier models could exceed 7 minutes by 2028 and 25 minutes by 2030, raising concerns about the effectiveness of current AI safety monitoring approaches.

This research addresses a critical vulnerability in current AI safety frameworks: the assumption that chain-of-thought reasoning can be monitored to ensure safe model behavior. If advanced models develop the ability to perform sophisticated reasoning internally without generating explicit thinking tokens, traditional oversight mechanisms become largely ineffective. The study's empirical measurement across 30,000+ questions spanning mathematics, coding, puzzles, and strategic reasoning provides concrete evidence of rapid capability progression rather than speculative concern.

The doubling of no-CoT task-completion time horizons annually reflects the broader acceleration in AI model capabilities over the past six years. This metric—representing the time a human would need to solve tasks that models complete with 50% accuracy—translates abstract capability improvements into human-relatable benchmarks. The projection that frontier models could reach 25-minute reasoning complexity by 2030 suggests that internal reasoning could eventually rival or exceed explicit CoT approaches in sophistication.

For the AI safety and governance communities, these findings expose a timing problem: oversight infrastructure designed around monitoring intermediate reasoning steps may become obsolete faster than alternative safeguards can be developed. The specific measurements of o3-mini reasoning token requirements provide quantifiable targets for tracking this capability drift. Developers of frontier models face mounting pressure to implement alternative transparency and control mechanisms beyond token monitoring, particularly as implicit reasoning capability approaches levels where meaningful human interpretation becomes impossible.

The research implicitly signals that current safety assumptions require urgent revision. Organizations deploying advanced AI systems should prepare for scenarios where traditional CoT-based auditing provides diminishing assurance, necessitating development of fundamentally different verification and alignment approaches before these capabilities fully mature.

• →Frontier AI no-CoT reasoning capability has doubled annually for six years, with GPT-5.5 exceeding 3-minute task-completion horizons.
• →Current AI safety monitoring approaches relying on explicit chain-of-thought tokens face obsolescence if models develop sophisticated implicit reasoning.
• →Projections estimate frontier models could reach 7+ minutes by 2028 and 25+ minutes by 2030 of internal reasoning complexity.
• →The study provides quantifiable metrics for tracking capability drift, enabling more precise monitoring of model evolution.
• →Developers must prioritize alternative safety and transparency mechanisms beyond CoT monitoring before implicit reasoning capabilities mature.