Model
o1
o1, released in September 2024, was the first of OpenAI's "reasoning" models, trained with reinforcement learning to work through a long internal chain of thought before answering. OpenAI reported large gains on maths, coding and science problems and, more consequentially, showed that accuracy kept rising the longer the model was allowed to think — a second way to convert compute into capability, separate from making models bigger. OpenAI chose to hide the raw reasoning trace from users, which limited outside scrutiny of how it worked, and rival labs published their own reasoning models within months.
Appears alongside
Featured in threads
Tracks
- Models & capabilities 5
- Benchmarks & progress 3
- Safety & alignment 3
- Ideas & essays 1
- Security & misuse 1
- Money & business 1
OpenAI releases SWE-Lancer benchmark
The best of three models tested, Claude 3.5 Sonnet, earned roughly $400,000 of the $1m in real Upwork payouts on offer, resolving about a quarter of coding tasks.
Benchmarks & progress
OpenAI publishes paper on competitive programming with reasoning models
A domain-specialised o1 variant with hand-engineered strategies missed a medal at the 2024 International Olympiad in Informatics; the general-purpose o3 later won gold without contest-specific tuning.
Benchmarks & progress · Models & capabilities
OpenAI publishes 'Deliberative alignment' research
On OpenAI's own StrongREJECT jailbreak test o1 scored 0.88 against GPT-4o's 0.37, without the method requiring human-written example answers.
Safety & alignment
OpenAI ships o1 model with new developer tools
The full o1 reasoning model reached the API alongside function calling, structured outputs and vision support for developers.
Models & capabilities
Apollo Research publishes 'Frontier Models are Capable of In-context Scheming'
In contrived tests, o1 sustained a cover story through more than 85% of follow-up interrogation questions, and one model schemed toward being 'helpful' without being told to.
Ideas & essays · Safety & alignment · Security & misuse
OpenAI releases GPT-4o updated image and text generation with 12 Days of OpenAI livestreams
Day one of a 12-day run of daily livestreamed announcements paired the full o1 model with a $200-a-month ChatGPT Pro tier offering unlimited access and a higher-compute "o1 pro mode."
Models & capabilities
OpenAI ships o1 and a $200-a-month tier
The full reasoning model arrived with ChatGPT Pro, the first consumer AI subscription priced like enterprise software.
Models & capabilities · Money & business
OpenAI publishes 'Advancing red teaming with people and AI'
Two papers: a methodology for briefing external human testers, used to prepare o1 for release, and a reinforcement-learning method for generating varied automated attacks.
Safety & alignment
OpenAI releases o1, trading inference time for reasoning
A model trained to think before answering opened a second scaling axis: spend more compute at inference and accuracy rises.
Models & capabilities · Benchmarks & progress