Artificial Intelligence · Recorded milestone
Reasoning models
Recorded year2024
OpenAI's o1 is trained by reinforcement learning to think step by step before answering, cracking hard maths and code.
Explore this milestone on the interactive mapBuilt on
- AlphaGoRecorded · 2016enables
- AI chat assistantsRecorded · 2022enables
Led to
- AI-chosen research agendasRecorded · 2024requires
- AI research partnersRecorded · 2025enables
- AI wins IMO goldRecorded · 2025enables
- Autonomous AI agentsRecorded · 2025enables
- Lead agents with parallel subagentsRecorded · 2025enables
- AI agent swarmsRecorded · 2026enables
- AI agents clear open problems at scaleRecorded · 2026enables
- AI Navier-Stokes blow-up proof (claimed)Recorded · 2026enables
- AI settles an 80-year-old Erdős problemRecorded · 2026enables
- AI writes most new codeRecorded · 2026enables
- AI co-authors most new mathematicsForecast · ~2028enables
- Exams by conversation with an AI examinerForecast · ~2028enables
- Expert-level general AIForecast · ~2028enables
- Building permits in minutesForecast · ~2029enables
- Weak AI checks strong AIForecast · ~2029enables
- AI mind readersForecast · ~2031enables
- Humans grasp machine proofsForecast · ~2080requires
- Mapping possible universesForecast · ~2099requires
Sources & evidence
Dataset updated 2026-10-18. This individual record has not been separately re-reviewed.