OpenAI introduced o1-preview and o1-mini on September 12, 2024, as the first preview models in a series designed to spend more time reasoning before responding. The launch emphasized math, coding, and science, but its benchmark results were OpenAI-reported evaluations—not guarantees for everyday prompts. The original access limits and feature gaps were specific to the 2024 launch; today, OpenAI marks o1 as deprecated in its API documentation.
What was OpenAI o1?
OpenAI described o1 as a new model series trained to work through complex tasks, with particular emphasis on science, coding, and mathematics. The company said the models spend additional time thinking before responding and suggested use cases such as scientific research and multi-step developer workflows. Those were OpenAI’s proposed applications, not independently verified outcomes for users.
In its technical explanation, OpenAI said it trained o1 with large-scale reinforcement learning and reported that performance improved with more training-time compute and more time spent thinking at inference. That description concerns the approach OpenAI reported; it does not establish that users can inspect a complete or transparent internal reasoning trace.
What OpenAI reported about o1’s performance
OpenAI highlighted benchmark and contest results to illustrate the models’ reasoning capabilities. These figures describe the company’s evaluations and should be read in the context of the named tests—not as predictions of results on every task.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
- Math: OpenAI said a reasoning model scored 83% on a qualifying exam for the International Mathematics Olympiad, compared with 13% for GPT-4o. The launch announcement explicitly attributed the 83% result to the “next model update” then in development; it should not be read as a result achieved by the released o1-preview.
- Coding: OpenAI said the model reached the 89th percentile in Codeforces competitions.
- Math contest performance: OpenAI’s technical post said o1 placed among the “top 500 students in the US” in a qualifier for the USA Math Olympiad (AIME). This is the company’s description of performance in that qualifier, not a direct ranking against all US students.
- Science: OpenAI reported that o1 exceeded human PhD-level accuracy on GPQA, a benchmark of physics, biology, and chemistry problems. That is a benchmark-specific claim, not evidence that the model has broad PhD expertise.
OpenAI also reported a score of 84 for o1-preview versus 22 for GPT-4o on one of its hardest jailbreak tests, using a 0–100 scale. This is a result on OpenAI’s test; it does not measure overall safety in every setting.
How o1-preview and o1-mini differed
At launch, OpenAI positioned the two models for different trade-offs. The descriptions below reflect the September 2024 announcement, not a current product recommendation.
| Model | Launch-era positioning | What that meant for users |
|---|---|---|
| o1-preview | Broader reasoning model for complex tasks | Aimed at work where deeper reasoning mattered more than speed or cost. |
| o1-mini | Smaller, faster, and cheaper; especially effective at coding | Intended for reasoning tasks that did not require broad world knowledge. |
What the launch preview could and could not do
The first ChatGPT preview lacked several features available in other ChatGPT experiences. OpenAI said it did not yet support web browsing or file and image uploads, and described those capabilities as future plans for the preview. OpenAI also said GPT-4o could be more capable for many common cases at that time, where broad world knowledge and those features mattered.
The initial API also lacked function calling, streaming, system-message support, and other features, according to the launch announcement. These are historical launch conditions, not a description of the current API.
How access worked at launch—and what changed
In September 2024, ChatGPT Plus and Team users could select o1-preview and o1-mini. OpenAI listed weekly limits of 30 messages for o1-preview and 50 for o1-mini. Enterprise and Edu access was planned for the following week. For the API, qualifying tier-5 developers could prototype at 20 requests per minute. Each of those limits and access terms applied to the launch period only.
OpenAI’s later documentation changes the present-day picture. Its current o1 API model page labels o1 “Deprecated,” and its technical post says o1-preview and o1-mini were retired from the API and o1 is deprecated. OpenAI’s changelog also records the release of o1-pro in March 2025. Model lifecycle information can change, so consult the live o1 API model page before making an access decision.
Rank #4
What OpenAI’s safety evaluations do—and do not—show
OpenAI’s o1 system card discusses evaluations covering harmfulness, jailbreak robustness, hallucinations, bias, chain-of-thought risks, and external red teaming. The company characterized the family as more robust on its hardest jailbreak evaluations, while also acknowledging that stronger reasoning capabilities could enable dangerous applications.
The system card cautions that results are scoped to the evaluated models and conditions: “The evaluations described in this System Card pertain to the full family of o1 models, and exact performance numbers for the model used in production may vary slightly depending on system updates, final parameters, system prompt, and other factors.” It also describes evaluations across multiple checkpoints, so its findings should not be treated as a blanket safety guarantee or assumed to describe every production configuration.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11What the o1 release meant
The September 2024 release introduced OpenAI’s reasoning-focused preview models and made the company’s ambitions—and its early trade-offs—visible: stronger performance on selected complex benchmarks, but missing features and launch-era access limits. The benchmark claims are useful context when attributed and kept within their test scope. For anyone considering the models now, the decisive practical point is that OpenAI’s API documentation marks o1 deprecated; the 2024 launch terms do not establish current availability.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
