OpenAI o1-preview was the first model in OpenAI’s o-series of reasoning models, released on September 12, 2024. It is retired: the full o1 model replaced it in ChatGPT on December 5, 2024, and OpenAI removed it from the API on July 28, 2025. For o1-style reasoning today, use GPT-6 Astra or GPT-6 Sol.
Last updated: September 2026
What was OpenAI o1-preview?
o1-preview was a research preview of a new kind of OpenAI model. OpenAI said the series was “designed to spend more time thinking before they respond” and could “reason through complex tasks and solve harder problems than previous models in science, coding, and math.” It launched in ChatGPT and the API on September 12, 2024, alongside a smaller, cheaper sibling, o1-mini.
The key idea was chain-of-thought reasoning learned through reinforcement learning. According to OpenAI, its “large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought.” Instead of answering at once, the model produced a long internal chain of thought, and OpenAI found that performance improved with more training compute and with more time spent thinking at answer time.
That made o1-preview matter well beyond its short life. It started the o-series (o1, o3, o4-mini) and established the “thinking” pattern that OpenAI later folded into GPT-5 and today’s GPT-6 models. According to Wikipedia, the project was known inside OpenAI first as Q* and later as Strawberry.
Is o1-preview still available in 2026?
No. The table below shows the full timeline, from launch to the retirement of the models that replaced it. Dates come from OpenAI’s announcements, help center and deprecations page.
| Date | What happened |
|---|---|
| September 12, 2024 | o1-preview and o1-mini launch in ChatGPT (Plus and Team) and in the API for usage tier 5 developers |
| September 17, 2024 | OpenAI raises ChatGPT limits from 30 to 50 messages a week for o1-preview, and to 50 a day for o1-mini |
| December 5, 2024 | o1 comes out of preview; paid ChatGPT users get it in the model selector, replacing o1-preview. ChatGPT Pro launches with o1 pro mode |
| December 2024 | o1 rolls out in the API to usage tier 5 developers |
| April 16, 2025 | o3, o4-mini and o4-mini-high replace o1, o3-mini and o3-mini-high in ChatGPT’s model selector |
| April 28, 2025 | OpenAI notifies developers that o1-preview and o1-mini will be removed from the API |
| July 28, 2025 | o1-preview removed from the API; recommended replacement: o3 |
| August 7, 2025 | GPT-5 becomes ChatGPT’s default, replacing o3 and o4-mini among others |
| October 27, 2025 | o1-mini removed from the API; recommended replacement: o4-mini |
| April 22, 2026 | OpenAI announces that o1, o1-pro, o3-mini and o4-mini will be removed from the API on October 23, 2026 |
| June 11, 2026 | OpenAI announces removal of the o3-2025-04-16 snapshot on December 11, 2026 |
| September 2026 | GPT-6 Astra launches (early September), followed by GPT-6 Sol and GPT-6 Luna on September 22 |
OpenAI’s o1-preview model page still describes it as a “preview of our first o-series reasoning model” and marks the snapshot o1-preview-2024-09-12 as deprecated. Any code that still calls o1-preview will fail and must be moved to a current model.
What could o1-preview do?
OpenAI positioned o1-preview for hard, multi-step problems rather than everyday chat. These are the capabilities it highlighted at launch.
Math, science and coding reasoning
OpenAI published these figures in its “Learning to reason with LLMs” research post. They are OpenAI’s own results, not independent tests:
| Benchmark (OpenAI’s figures) | GPT-4o | o1-preview | o1 |
|---|---|---|---|
| AIME 2024 math exam, pass@1 / consensus@64 | 9.3% / 13.4% | 44.6% / 56.7% | 74.4% / 83.3% |
| Codeforces rating (percentile) | 808 (11th) | 1,258 (62nd) | 1,673 (89th) |
| GPQA Diamond science, pass@1 | 50.6% | 73.3% | 77.3% |
The widely quoted “83%” math score and “89th percentile” coding result from the launch belong to the full o1 model, which OpenAI previewed at the same time, not to o1-preview itself.
Hidden chain of thought
OpenAI decided “not to show the raw chains of thought to users” and showed a model-generated summary instead. In the API, reasoning tokens were invisible but still counted toward the context window and were billed as output tokens.
Safety through reasoning
OpenAI said o1-preview could reason about its safety rules in context. On one of its hardest jailbreaking tests, OpenAI reported that GPT-4o scored 22 out of 100 while o1-preview scored 84.
What it lacked
At launch, OpenAI said o1-preview did not yet have “many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images.” OpenAI also noted it was not preferred on some natural-language tasks, so GPT-4o stayed the better choice for many everyday requests. It was slower and more expensive, and the full o1 later added image input, while the o1 API added function calling, Structured Outputs, developer messages and a reasoning_effort setting.
How much did o1-preview cost, and what does reasoning cost now?
o1-preview cannot be bought anymore, so its prices are historical. At launch, ChatGPT Plus and Team users got 30 o1-preview messages a week, later raised to 50, and API access was limited to tier 5 developers at 20 requests per minute. Current prices below are from OpenAI’s pricing and model pages as of September 2026, in USD per 1 million tokens:
| API model | Status | Input | Output |
|---|---|---|---|
| o1-preview (historical) | Removed July 28, 2025 | $15.00 | $60.00 |
| o1 | Removal on October 23, 2026 | $15.00 | $60.00 |
| o3 | Snapshot o3-2025-04-16 removal on December 11, 2026 | $2.00 | $8.00 |
| gpt-6-astra | Current flagship | $10.00 | $50.00 |
| gpt-6-sol | Current (September 22, 2026) | $2.00 | $10.00 |
| gpt-6-luna | Current (September 22, 2026) | $0.10 | $0.50 |
| gpt-5.6-sol | Current; OpenAI’s listed substitute for o1 | $4.00 | $20.00 |
| gpt-5.6-terra | Current | $2.00 | $12.00 |
The comparison is stark: GPT-6 Sol costs less than a seventh of o1-preview’s input price and one sixth of its output price. OpenAI says GPT-5.6 Sol’s promotional pricing lasts at least through November 21, 2026, and GPT-6 Astra and Sol charge 2x input and 1.5x output rates for requests above 272K input tokens. Reasoning tokens are still billed as output on every current model.
In ChatGPT, plans in September 2026 are Free, Go ($8 per month in the US), Plus ($20 per month), two Pro tiers (Pro $100 with 5x Plus usage and Pro $200 with 20x), plus Business and Enterprise. OpenAI paused new sign-ups and upgrades to Pro $200 on September 10, 2026; existing subscribers are not affected.
What replaced o1-preview?
Replacement came in several steps, and every model in the chain has since been replaced or scheduled for removal:
- o1 (December 2024): the finished model. OpenAI’s o1 system card calls it “the next model in this series (previously OpenAI o1-preview).” It has a 200,000-token context window, up from 128,000.
- o3 and o4-mini (April 2025): OpenAI named o3 as the API replacement for o1-preview and o4-mini for o1-mini.
- GPT-5 (August 2025): became ChatGPT’s default model, with “GPT-5 Thinking” still selectable in the model picker for paid users.
- GPT-5.6 Sol and Terra (2026): the substitutes OpenAI’s deprecations page lists for o1, o1-pro, o3, o3-pro, o3-mini (all gpt-5.6-sol) and o4-mini (gpt-5.6-terra).
- GPT-6 Astra, Sol and Luna (September 2026): OpenAI’s current family. OpenAI says Sol and Luna upgrade their GPT-5.6 counterparts at half the API price.
How to get o1-style reasoning today
- In ChatGPT, use a paid plan. OpenAI says GPT-6 Astra is rolling out to Plus, Pro, Business and Enterprise users. GPT-6 Sol and Luna launched in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and Free and Go users can try GPT-6 Luna in the desktop app.
- For the hardest problems, pick Pro. OpenAI’s pricing page lists “Pro reasoning powered by GPT-6 Astra” for Pro subscribers, and OpenAI says Pro, Business and Enterprise users get GPT-6 Astra Pro, the closest current equivalent to the o1 pro mode that Pro launched with in 2024.
- In the API, replace the model name. OpenAI’s reasoning guide says to start with gpt-6-astra for most reasoning workloads. OpenAI’s model pages describe gpt-6-sol as built for complex coding and agentic workflows and gpt-6-luna as its most efficient model for high-volume tasks.
- Set reasoning effort. GPT-6 Sol and Luna accept effort values of none, low, medium (the default), high, xhigh and max; GPT-6 Astra runs from low to max. o1-preview had no such control. Start at medium and raise it only if quality needs it.
- Migrate before the deadlines. If you moved from o1-preview to o1, o3-mini or o4-mini, you must move again before October 23, 2026; o3 users have until December 11, 2026.
o1-preview alternatives from other AI labs
- Claude (Anthropic): pick it for long agentic work and writing. Claude Fable 5.1 ($10/$50 per million tokens) is Anthropic’s model “for demanding reasoning,” and Claude Opus 5.5 ($4/$20) and Claude Sonnet 5 ($2/$10) use adaptive thinking with 1M-token context.
- Gemini (Google): pick it if you work in Google’s ecosystem. Gemini 3.1 Pro (preview) costs $2 input and $12 output per million tokens up to 200K tokens, and Gemini 3.8 Flash costs $0.75 and $3.75 through December 31, 2026; Google includes thinking tokens in output prices.
- DeepSeek: pick it for the lowest reasoning costs. Its current API models, deepseek-v4-pro and deepseek-flash, run in thinking mode by default with 1M-token context, and prices vary by peak and off-peak hours, with deepseek-flash output from $0.60 per million tokens off-peak.
Frequently asked questions
Is OpenAI o1-preview still available?
No. OpenAI replaced o1-preview in ChatGPT with the full o1 model on December 5, 2024, and removed o1-preview from its API on July 28, 2025, after notifying developers on April 28, 2025. The API model page still exists for reference, but you cannot call the model. OpenAI’s deprecations page named o3 as the replacement at the time.
When was o1-preview released?
OpenAI released o1-preview, together with o1-mini, on September 12, 2024. It was available in ChatGPT to Plus and Team users, with Enterprise and Edu access the following week, and in the API to usage tier 5 developers with a limit of 20 requests per minute. OpenAI called it the first model in its o-series of reasoning models.
What replaced o1-preview?
The full o1 model replaced it in ChatGPT on December 5, 2024, and OpenAI recommended o3 as its API replacement. Both were later superseded too: o3 and o4-mini replaced o1 in ChatGPT in April 2025, and GPT-5 replaced o3 as ChatGPT’s default in August 2025. In September 2026, OpenAI’s current models are GPT-6 Astra, GPT-6 Sol and GPT-6 Luna.
How much did o1-preview cost?
In the API, o1-preview cost $15 per million input tokens, $7.50 per million cached input tokens and $60 per million output tokens, according to its OpenAI model page. Hidden reasoning tokens were billed as output tokens. In ChatGPT it was included with Plus and Team, with weekly message caps. These prices are historical because the model is no longer sold.
Is the o1 model also being shut down?
Yes. On April 22, 2026, OpenAI announced that o1, o1-pro, o3-mini and o4-mini will be removed from the API on October 23, 2026. It recommends gpt-5.6-sol as the substitute for o1, and gpt-5.6-sol with reasoning mode set to pro for o1-pro. The o3 snapshot o3-2025-04-16 follows on December 11, 2026.
Which OpenAI model should I use for reasoning in 2026?
OpenAI’s reasoning guide recommends starting with gpt-6-astra for most reasoning workloads, at $10 per million input tokens and $50 per million output tokens. OpenAI positions GPT-6 Sol ($2 and $10) for coding and agentic work and GPT-6 Luna ($0.10 and $0.50) for high-volume tasks. In ChatGPT, reasoning models come with paid plans starting with Plus at $20 per month.
Did o1-preview show its chain of thought?
No. OpenAI chose not to show the raw chain of thought to users. It said it would instead show a model-generated summary of the reasoning. In the API, reasoning tokens stay hidden but still use context space and are billed as output tokens, a design that OpenAI’s current reasoning models keep, with optional reasoning summaries.
Verdict
o1-preview was a turning point. It showed that letting a model think before answering could lift math, science and coding results well above GPT-4o by OpenAI’s own measures, and it started the o-series that shaped every OpenAI model since. It was also slow, expensive for its time and missing basic ChatGPT features, which is why OpenAI replaced it within three months.
In September 2026, o1-preview is gone from ChatGPT and the API, and even its successors o1, o3-mini and o4-mini leave the API on October 23, 2026. For reasoning in ChatGPT, use a paid plan with GPT-6 Astra. Developers should start with gpt-6-astra, or gpt-6-sol for a lower price, and compare Claude, Gemini and DeepSeek on their own tasks.
Sources
- Deprecations (OpenAI API docs)
- o1-preview model page (OpenAI API docs)
- o1 model page (OpenAI API docs)
- Introducing OpenAI o1-preview (OpenAI)
- Learning to reason with LLMs (OpenAI)
- Introducing ChatGPT Pro (OpenAI)
- OpenAI o1 System Card (OpenAI)
- OpenAI releases o1 LLM, unveils ChatGPT Pro (InfoWorld)
- OpenAI o1 and new tools for developers (OpenAI)
- Introducing OpenAI o3 and o4-mini (OpenAI)
- Introducing GPT-5 (OpenAI)
- Deprecation notice: upcoming model shutdowns in 2026 (OpenAI Developer Community)
- API pricing (OpenAI)
- Models (OpenAI API docs)
- Reasoning models guide (OpenAI API docs)
- GPT-6 Astra model page (OpenAI API docs)
- GPT-6 Sol model page (OpenAI API docs)
- GPT-6 Luna model page (OpenAI API docs)
- GPT-5.6 Sol model page (OpenAI API docs)
- GPT-5.6 Terra model page (OpenAI API docs)
- GPT-6 Astra: A new generation of intelligence (OpenAI)
- Introducing GPT-6 Sol and Luna (OpenAI)
- Announcing GPT-6 Sol and GPT-6 Luna in the API, Codex and ChatGPT (OpenAI Developer Community)
- OpenAI launches GPT-6 Sol and Luna (TechCrunch)
- ChatGPT pricing (OpenAI)
- What is ChatGPT Plus? (OpenAI Help Center)
- About ChatGPT Pro tiers (OpenAI Help Center)
- Introducing ChatGPT Go, now available worldwide (OpenAI)
- Model release notes (OpenAI Help Center)
- OpenAI o1 (Wikipedia)
- Models overview (Claude Platform docs)
- Pricing (Claude Platform docs)
- Gemini models (Gemini API docs)
- Gemini API pricing (Google AI for Developers)
- Models and pricing (DeepSeek API docs)
