OpenAI Reasoning Models

OpenAI's reasoning models, the "o" series from o1-preview (Sep 12, 2024) through o3 and o4-mini (Apr 16, 2025), think before answering. GPT-5 merged them into its main models in August 2025, and today reasoning is a level you pick in ChatGPT (Medium to Extra High) or an effort setting in the API.

OpenAI's reasoning models are models trained to work through a problem before answering: the "o" series ran from o1-preview on Sep 12, 2024 to o3 and o4-mini on Apr 16, 2025. Since GPT-5 on Aug 7, 2025, OpenAI has no separate reasoning line: its main models think when asked, and in ChatGPT you choose how hard with the Medium, High and Extra High levels. Every o-series model has left ChatGPT, and the last of them leave the API on Dec 11, 2026.

Quick facts

First	o1-preview and o1-mini, Sep 12, 2024
Last	o3 and o4-mini, Apr 16, 2025; o3-pro Jun 10, 2025
Merged into	GPT-5, Aug 7, 2025
Left ChatGPT	o4-mini Feb 13, 2026; o3 Aug 26, 2026 (scheduled)
API shutdown	o1, o1-pro, o3-mini, o4-mini Oct 23, 2026; o3, o3-pro Dec 11, 2026
Reasoning today	ChatGPT levels Medium, High, Extra High, Pro; API effort none to max

The o-series, model by model

Model	Released	Notes
o1-preview, o1-mini	Sep 12, 2024	Plus and Team; 30 and 50 messages a week at launch, raised on Sep 17, 2024 to 50 a week and 50 a day
o1, o1 pro mode	Dec 5, 2024	Came with the first $200 ChatGPT Pro plan, which included unlimited o1
o3-mini	Jan 31, 2025	Offered on Free as well as Plus, Pro and Team
o3, o4-mini	Apr 16, 2025	o3 also powered codex-1, the first Codex cloud agent's model (OpenAI)
o3-pro	Jun 10, 2025	Replaced o1-pro for Pro and Team

The models also powered research tools: o3-deep-research and o4-mini-deep-research were API models until Jul 23, 2026.

How reasoning merged into GPT-5 and GPT-6

3:08 GPT-5 folded the o-series in. OpenAI described it as a fast model, "GPT-5 thinking", and a router that chooses between them, and named it the replacement for o3 and o4-mini, with GPT-5 pro replacing o3-pro. OpenAI claimed GPT-5 thinking made about 80% fewer factual errors than o3.

From there, reasoning became a dial rather than a model:

Nov 18, 2025: the Extra High (xhigh) effort arrived in Codex with GPT-5.1-Codex-Max, and in the API with GPT-5.2 on Dec 11, 2025.
Mar 17, 2026: ChatGPT's picker became Instant, Thinking and Pro, with thinking effort under Configure.
Jun 10, 2026: the levels were renamed Medium (Thinking Standard), High (Thinking Extended) and Extra High (Thinking Heavy), and Thinking Light was removed.
Sep 14, 2026: ChatGPT stopped switching from Instant to Thinking by itself for Plus and Pro.

Today the Chat tab's Medium to Extra High levels run on GPT-6 Sol, and Pro on GPT-6 Astra (OpenAI Help Center). Plus gets Medium and High; Pro, Business and Enterprise get all of them. Free and Go have a Think button instead. See ChatGPT Models.

In the API, the GPT-6 models take a reasoning effort of low, medium, high, xhigh or max, and Terra and Luna also accept none (OpenAI API docs). Since GPT-6, effort can even be changed mid-conversation without losing the prompt cache. Codex calls its levels Light, Medium, High, Extra High, Max and Ultra, where Ultra runs subagents in parallel.

Retirements

In ChatGPT, GPT-5 replaced o3 and o4-mini as the default in August 2025, but both stayed selectable for a while. o4-mini left on Feb 13, 2026. On May 28, 2026 OpenAI said "OpenAI o3 will be retired from ChatGPT on August 26, 2026 following a 90-day sunset period", and the o3 launch page now says o3 and o4-mini "are no longer available in ChatGPT".

In the API: o1-preview ended Jul 28, 2025 and o1-mini Oct 27, 2025; o1, o1-pro, o3-mini and o4-mini shut down on Oct 23, 2026; o3 and o3-pro shut down on Dec 11, 2026, with gpt-5.6-sol named as the replacement (o3-pro maps to GPT-5.6 Sol with reasoning.mode: pro) (OpenAI API docs). Until then o3 costs $2 input and $8 output per million tokens.

Frequently asked questions

Can I still use o3 in ChatGPT? No. OpenAI scheduled o3's retirement from ChatGPT for Aug 26, 2026, and its launch page now says o3 is no longer available there. Use Medium or High in the model picker instead.

What replaced OpenAI o1 and o3? GPT-5 replaced them in August 2025 by building reasoning into the main model. Today the reasoning levels in ChatGPT run on GPT-6 Sol and GPT-6 Astra.

When do o3 and o3-pro shut down in the API? On Dec 11, 2026. OpenAI's replacement is gpt-5.6-sol, with its pro mode for o3-pro.

What is the difference between Instant and Thinking in ChatGPT? Instant answers at once; the thinking levels (Medium, High, Extra High) spend more time reasoning before the reply. OpenAI renamed Thinking Standard, Extended and Heavy to Medium, High and Extra High on Jun 10, 2026.