OpenAI Models in 2026: Current ChatGPT, API, Codex, and Retired Models

OpenAI’s 2026 model lineup is easiest to understand by separating three things that are often confused: what appears in ChatGPT, what is documented for developers in the API, and what has been retired from ChatGPT. The sources here are OpenAI model docs, plan-limit pages, release notes, and retirement notices, so this should be treated as a verified availability snapshot rather than a complete launch-history timeline.

The 2026 lineup at a glance

Category	Verified model names in the cited docs	What it means
ChatGPT reasoning	GPT-5.4 Thinking; GPT-5.4 / GPT-5.2 Thinking in plan docs	OpenAI describes GPT-5.4 Thinking as its most capable reasoning model in ChatGPT. Enterprise/Edu plan docs list a Thinking row with a long-term limit of 200 per week and a temporary increase above that level.
ChatGPT general use	GPT-5.3 / GPT-5.2 Instant	Enterprise/Edu model-limit docs list this Instant row with unlimited usage.
ChatGPT Pro	GPT-5.4 / GPT-5.2 Pro	Enterprise/Edu docs list this Pro row with 15 requests per month. A release note also says existing GPT-5.1 Pro conversations continue with GPT-5.4 Pro.
ChatGPT mini or fallback	GPT-5.4 mini	OpenAI says GPT-5.4 mini began rolling out in ChatGPT on March 18, 2026, with Free and Go access through the “Thinking” feature and fallback access for other users when rate limits apply.
API image, realtime voice, and speech	GPT Image 1.5 / gpt-image-1.5, gpt-image-1-mini, gpt-realtime-1.5, gpt-realtime-mini, gpt-4o-mini-tts	OpenAI’s API model docs list these as specialized models for image generation/editing, realtime speech-to-speech, and text-to-speech.
Codex	GPT-5.4, GPT-5.4-Mini, GPT-5.3-Codex, GPT-5.2-Codex, GPT-5.2, GPT-5.1-Codex-Max, GPT-5.1-Codex-mini	OpenAI’s Codex rate card lists these models and says code review uses GPT-5.3-Codex.
Retired from ChatGPT	GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, GPT-5 Instant/Thinking, GPT-5.1 Instant/Thinking/Pro	OpenAI retirement notices say these models were removed from ChatGPT in 2026, with API access treated separately.

Current ChatGPT models documented for 2026

The highest-end ChatGPT model in the cited documentation is GPT-5.4 Thinking. OpenAI describes it as its most capable reasoning model in ChatGPT and says it is designed for difficult real-world work, including documents, images, tools, research, hard math, spreadsheets, frontend code, and slide creation.

For Enterprise and Edu plans, OpenAI’s model-limit page lists three main rows: GPT-5.3 / GPT-5.2 Instant with unlimited usage, GPT-5.4 / GPT-5.2 Thinking with a long-term limit of 200 per week, and GPT-5.4 / GPT-5.2 Pro with 15 requests per month. Because this is plan-limit documentation, it should not be read as a guarantee that every ChatGPT plan or workspace will show the same picker, naming, or usage limits.

GPT-5.4 mini is also documented for ChatGPT. OpenAI’s Enterprise and Edu release notes say it started rolling out on March 18, 2026; Free and Go users can access it through the “Thinking” feature in the plus menu, while other users may see it as a rate-limit fallback.

ChatGPT and API availability are different

A model leaving ChatGPT does not automatically mean it has left the API. OpenAI’s pricing page says OpenAI APIs are billed separately from ChatGPT Plus, Business, Enterprise, and Edu subscriptions. OpenAI’s retirement announcement for GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini also said there were no API changes at that time.

That distinction matters for developers, because the API catalog includes specialized models that are not simply the same as the ChatGPT model picker. In the cited API model docs, OpenAI lists image models, realtime speech-to-speech models, and a text-to-speech model.

Developer use case	Models listed in the cited API docs	Source-backed note
Image generation and editing	GPT Image 1.5 / gpt-image-1.5, gpt-image-1-mini	OpenAI describes GPT Image 1.5 as a state-of-the-art image generation model and gpt-image-1-mini as a cost-efficient version of GPT Image 1.
Realtime voice	gpt-realtime-1.5, gpt-realtime-mini	OpenAI lists gpt-realtime-1.5 as its best voice model for audio input and output, and gpt-realtime-mini as a cost-efficient realtime model.
Text-to-speech	gpt-4o-mini-tts	OpenAI lists GPT-4o mini TTS for generating natural-sounding speech from text.
Codex work	GPT-5.4, GPT-5.4-Mini, GPT-5.3-Codex, GPT-5.2-Codex, GPT-5.2, GPT-5.1-Codex-Max, GPT-5.1-Codex-mini	OpenAI’s Codex rate card lists these models and says code review uses GPT-5.3-Codex.

Models retired from ChatGPT in 2026

Several familiar OpenAI model names are no longer current ChatGPT choices in the cited OpenAI retirement notices.

Date	Retired from ChatGPT	Important exception or API note
February 13, 2026	GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant/Thinking	OpenAI’s retirement announcement said there were no API changes at that time.
March 11, 2026	GPT-5.1 Instant, GPT-5.1 Thinking, and GPT-5.1 Pro	OpenAI says existing GPT-5.1 conversations continue with GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro, depending on the original model.
April 3, 2026	GPT-4o exception window ended for Custom GPTs	OpenAI says ChatGPT Business, Enterprise, and Edu customers retained access to GPT-4o within Custom GPTs until April 3, 2026; after that, GPT-4o would be fully retired across all plans.

Older model names you may still see

Some older names still matter because they appear in announcements, API discussions, or retirement notices.

GPT-4.1 was introduced as an API family with GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. OpenAI described that family as improving coding and instruction following and supporting up to 1 million tokens of context. GPT-4.1 and GPT-4.1 mini were later retired from ChatGPT on February 13, 2026, according to OpenAI’s retirement notice.

o3 and o4-mini were introduced as o-series reasoning models trained to think longer before responding. OpenAI said that, in ChatGPT, these models could use tools such as web search, file analysis, Python, visual reasoning, and image generation. OpenAI later included o4-mini in the February 13, 2026 group retired from ChatGPT.

OpenAI also documents open models named gpt-oss-120b and gpt-oss-20b. One OpenAI page says gpt-oss-120b reaches near-parity with OpenAI o4-mini on core reasoning benchmarks while running efficiently on a single 80 GB GPU. Those sources establish the gpt-oss model names, but they should not be treated as a ChatGPT model-picker list or a complete API availability list.

Which OpenAI model should you use in 2026?

• For the hardest ChatGPT reasoning, research, math, document, image, and tool-heavy work, start with GPT-5.4 Thinking when your plan offers it.
• For broad Enterprise/Edu ChatGPT usage, the cited plan table lists GPT-5.3 / GPT-5.2 Instant with unlimited usage.
• For lighter ChatGPT access or fallback behavior, GPT-5.4 mini is the documented 2026 mini model in the cited release notes.
• For image generation or image editing through the API, use the API model docs for GPT Image 1.5 / gpt-image-1.5 and gpt-image-1-mini.
• For realtime voice or text-to-speech API work, the cited API docs list gpt-realtime-1.5, gpt-realtime-mini, and gpt-4o-mini-tts.
• For Codex-related work, use the Codex rate card; it lists GPT-5.4, GPT-5.3-Codex, GPT-5.2-Codex, GPT-5.2, and GPT-5.1-Codex variants, and says code review uses GPT-5.3-Codex.

Bottom line

The most defensible shorthand is this: OpenAI’s cited 2026 ChatGPT documentation points to GPT-5.4 Thinking/Pro, GPT-5.3 Instant, GPT-5.4 mini, and GPT-5.2 variants in plan rows, while the API and Codex documentation separately list specialized image, realtime, speech, and coding models. GPT-4o, GPT-4.1, GPT-4.1 mini, o4-mini, GPT-5 Instant/Thinking, and GPT-5.1 variants should be treated as retired from ChatGPT rather than current ChatGPT model choices.