Updated October 24, 2026. Historical coverage of the September 2024 launch; original publication date retained.
Availability update: OpenAI’s API deprecation schedule lists o1-preview as shut down on July 28, 2025, o1-mini on October 27, 2025, and the later o1/o1-2024-12-17 model on October 23, 2026. See the official API deprecation schedule and current model catalog before starting a new integration.
OpenAI’s September 2024 reasoning-model launch
OpenAI introduced o1-preview and o1-mini on September 12, 2024. This article covers that launch and its reported reasoning results, rather than recommending the models for current use. The original announcement described models trained to spend more computation reasoning before answering.
A New Paradigm in AI Reasoning
The o1 series represents a paradigm shift in AI development, prioritizing a more deliberate, thoughtful approach to problem-solving. These models, akin to human cognition, are trained to ‘think’ before responding, simulating the process of refining thought, exploring diverse strategies, and learning from errors.
Reported Performance in Complex Domains
OpenAI reported research o1 results alongside the released o1-preview; the two should not be treated as the same evaluated model. Its September 2024 technical report gave research o1 an AIME score of about 74% with one sample per problem and 83% using consensus across 64 samples. The latter is not a single-answer accuracy score. The report placed o1 at the 89th percentile on Codeforces and above the recruited PhD experts on GPQA-diamond questions in physics, chemistry and biology. These were bounded benchmark comparisons, not evidence of general equivalence to a PhD or guaranteed real-world accuracy.
Early Preview with Promising Potential
At launch, o1-preview lacked web browsing and file or image uploads in ChatGPT. OpenAI said GPT-4o remained more capable for many common tasks at that time. These are September 2024 product limitations, not a description of current ChatGPT.
The release introduced the o1 name for a reasoning-focused model series.
Safety as a Paramount Concern
OpenAI has implemented a novel safety training approach that leverages o1’s reasoning capabilities to improve adherence to safety and alignment guidelines. By enabling o1 to reason about safety rules contextually, it can apply them more effectively.
OpenAI measures safety through various methods, including testing how well models resist attempts to bypass safety rules, known as ‘jailbreaking’. In one of the most challenging jailbreaking tests, GPT-4o scored 22 (on a scale of 0-100), while the o1-preview model scored a significantly higher 84. These figures were OpenAI’s launch-time test results, not a guarantee that either model could not be jailbroken.
Reinforced Safety Measures
To complement the enhanced capabilities of these models, OpenAI has fortified its safety protocols, internal governance, and collaboration with federal governments. These measures include rigorous testing and evaluations using their Preparedness Framework, best-in-class red teaming, and board-level review processes, including oversight by their Safety & Security Committee.
Collaboration for AI Safety
OpenAI has formalized agreements with the U.S. and U.K. AI Safety Institutes, granting them early access to a research version of the o1 model. This collaborative effort aims to establish a robust process for research, evaluation, and testing of future models, both before and after their public release.
Applications Across Diverse Fields
OpenAI proposed research uses such as annotating cell-sequencing data, developing quantum-optics formulas and building multi-step developer workflows. These were illustrative applications in the launch announcement, not independently verified deployments or a substitute for expert review.
OpenAI o1-mini: A Cost-Effective Alternative
Alongside o1-preview, OpenAI released o1-mini, a smaller reasoning model aimed at coding and STEM tasks. Its announced API cost was 80% lower than o1-preview at launch. That historical comparison is not a current price or availability offer.
Access and Availability
On September 12, 2024, ChatGPT Plus and Team users began receiving o1-preview and o1-mini in the model picker, initially limited to 30 and 50 messages per week respectively. OpenAI’s September 17 update increased these to 50 per week for o1-preview and 50 per day for o1-mini.
At launch, Enterprise and Edu access was announced for the following week. API prototyping was initially limited to usage-tier-5 developers at 20 requests per minute. OpenAI also announced an intention to extend o1-mini to free users; that was a plan, not an access entitlement. None of these launch limits should be used to choose a current subscription.
Future Developments
The September 2024 announcement described plans for model updates, browsing and file/image uploads, alongside continued GPT-series development. Those statements recorded the roadmap at launch, not promises of future support for the now-retired preview model.
Conclusion
The o1 launch marked OpenAI’s shift toward allocating more computation to reasoning before answering. Its historical benchmarks help explain that development, but model identity, evaluation settings and later retirement dates matter when interpreting the results. For a new project, check current official model availability rather than relying on the launch access instructions above.
