ChatGPT versions: all 30 models from GPT-3.5 to GPT-6, with dates

Every ChatGPT version in order, with release and retirement dates

OpenAI has shipped eight ChatGPT model releases in the 14 months from GPT-5 onward, and only 11 in the 32 months before that. We read all of its ChatGPT release notes, its model release notes and each of the launch posts on 18 October 2026, and then we counted them. The orange dots are the models that you can still pick in ChatGPT today.

Each row is a model that OpenAI let ChatGPT users pick, or that it set as the default, in the order they came out. "Left ChatGPT" is the day OpenAI says it stopped being available in the app, which is not the same as the API. We left out the fallback models that never showed up in the picker, such as GPT-5 Thinking mini. GPT-1 (June 2018), GPT-2 (February 2019) and GPT-3 (May 2020) came before ChatGPT and never ran in it, so they are not rows either. All of the dates were read on 18 October 2026.

Two of the rows carry a caveat. OpenAI's release note from 7 October says that Free and Go "use GPT-6 Luna", but its model help page still named GPT-5.6 Luna as their default when we read it on the same day. And the o3 date is the one that OpenAI set for its retirement back on 28 May. No later note has moved it.

Which ChatGPT version is the latest?

GPT-6 is the latest ChatGPT version, and it reached the Chat tab on 7 October 2026. OpenAI's launch post says it's "powered by GPT-6 Sol for Plus, Pro, Business, and Enterprise tiers, and GPT-6 Luna for Free and Go tiers." When you pick the Pro level, it runs on GPT-6 Astra instead. So one picker holds three models.

This is what each plan gets in the Chat tab, according to OpenAI's help page on GPT-6 and the other models, which we read on 18 October 2026.

Two more models sit outside of the Chat tab. GPT-6.1 Sol came out on 29 September 2026, but it's only in ChatGPT Work and Codex. That's also where you'll find Terra and Luna from the GPT-5.6 family, and the help page says that they "are not selectable in standard ChatGPT conversations." If your picker doesn't show any of this yet, the rollout may not have reached your account.

How long does a ChatGPT model last before OpenAI retires it?

A ChatGPT model now lasts for about six months before it's retired. That's short. The eight retired models from GPT-5 onward stayed for between 112 and 196 days, with a median of 174. The 15 that came before GPT-5 had a median of 275 days, and GPT-4 held on for 778 of them. Across all 23 of the retired models, the median is 187 days. GPT-4o and o3 left after GPT-5 arrived, but they launched before it, so they count in the older group.

OpenAI wrote down its rule when it retired GPT-5.2 in June 2026: "Models will generally remain available in ChatGPT for 90 days after a successor is released." That's a floor, not a deadline. OpenAI had said that the Instant version of GPT-5.2 would go on 3 June and that the Thinking version would go on 5 June, yet both of them stayed until 12 June.

Old models don't vanish all at once. They move to a legacy list for paid users first. When GPT-5 came out on 7 August 2025, it replaced GPT-4o, o3, o4-mini, GPT-4.1 and GPT-4.5 for everyone who was signed in. Five days later, OpenAI put 4o back in the picker for paid users after they complained. It finally left on 13 February 2026, which was 641 days after its launch.

What happened to GPT-4o?

GPT-4o was retired from ChatGPT on 13 February 2026. On the same day, OpenAI also retired GPT-4.1, its mini version, o4-mini and both of the GPT-5 modes, and GPT-5 Pro followed on 19 February. The retirement notice says that all of these models "will continue to be available in the API."

That's how every retirement on this page works. A model leaves the ChatGPT app, but developers can keep calling it through the OpenAI API until OpenAI gives them a separate notice. It was the same for GPT-4, which left ChatGPT on 30 April 2025 and stayed in the API.

What do Sol, Luna, Terra and Astra mean in ChatGPT?

Sol, Luna, Terra and Astra are the size tiers within one GPT generation. OpenAI calls Terra "a balance of capability, speed, and cost for everyday work" and it calls Luna "the fastest and lowest-cost model in the GPT-5.6 family." Sol is the flagship reasoning model. Astra sits above it as the model behind GPT-6 Pro.

The names began with GPT-5.6 on 9 July 2026, which came out as three models at once. Before that, OpenAI split each generation by mode instead: Instant for the fast answers, Thinking for reasoning, and Pro for the hardest work. The picker still uses levels like these. On 10 June 2026 they were renamed to Instant, Medium, High and Extra High.

Can you still use a retired ChatGPT model?

Not inside of ChatGPT itself. Once OpenAI retires a model from the app, all of your old conversations with it move to its successor. On 11 March 2026, for example, the chats that were on GPT-5.1 moved to GPT-5.3 Instant, GPT-5.4 Thinking or GPT-5.4 Pro. The API is the only way that you can get back to the model itself.

Whizi Pro, which costs $29.99 a month, keeps several of the retired ChatGPT models next to the current ones, including GPT-4o, GPT-4.1, GPT-5.2 and GPT-5.5 (plans). What you get there is the model, not the features of the ChatGPT app such as Intelligent UI or Codex. If all you want is today's GPT-6 inside ChatGPT, then Plus at $20 a month is the more direct way to get it.