GPT-4.5: What It Was, When OpenAI Retired It and What Replaced It

Short answer: no, GPT-4.5 isn't available anymore. OpenAI took it out of the API on July 14, 2025, and out of ChatGPT on June 26, 2026, custom GPTs included. Old chats that used it carry on with a newer model. If you open ChatGPT today you get GPT-6: GPT-6 Luna on Free and Go, GPT-6 Sol on Plus, Pro, Business and Enterprise.

That's the answer most people come here for. If you want the rest of the story (what GPT-4.5 was good at, why such a big model lasted so little, and what to use instead), here it is.

Is GPT-4.5 still available? The timeline

GPT-4.5 had a short life by OpenAI's standards: about 16 months in ChatGPT, and less than five in the API. These are the dates as OpenAI published them.

What made GPT-4.5 different

GPT-4.5 was OpenAI's bet on scale. OpenAI called it its "largest and best model for chat yet", built by scaling up computing power and data, not by teaching it to think step by step before answering. OpenAI's own launch post was clear about it: "GPT‑4.5 doesn't think before it responds", which set it apart from reasoning models like o1. A couple of weeks before launch, Sam Altman had already called it "our last non-chain-of-thought model".

What that bought you was knowledge and a more natural tone, not puzzle-solving. Where OpenAI showed the difference most clearly was in how often it made things up. On SimpleQA, OpenAI's test of short factual questions, these were the numbers it published:

So it was wrong a lot less often than GPT-4o on that test, but still wrong on more than a third of the questions. To me, that's the honest way to read it: better, not trustworthy.

At launch it could search the web, take file and image uploads and work in canvas, but it didn't support Voice Mode, video or screen sharing in ChatGPT. If you mostly talked to ChatGPT out loud, GPT-4.5 was never your model.

OpenAI's own launch livestream from February 27, 2025, the day GPT-4.5 arrived for Pro users. It's the best way to hear how OpenAI itself pitched the model back then, so keep in mind it describes a model you can no longer pick in ChatGPT.

13:19

Why OpenAI let it go so soon

The short version: it was huge and expensive, and newer models did the job for less. In the API, GPT-4.5 cost $75 per million input tokens and $150 per million output tokens. When OpenAI launched GPT-4.1 in April 2025, it announced in the same breath that GPT-4.5 Preview would be turned off three months later, and pointed developers to GPT-4.1 instead.

In ChatGPT it hung on for almost a year longer, until OpenAI retired it in June 2026. By then the conversation had moved on to models that do reason step by step, which is exactly what GPT-4.5 was built not to do.

What to use instead today

If what you liked about GPT-4.5 was having a pleasant, well-informed chat, the honest answer is the default model in ChatGPT right now. Since early October 2026, that's GPT-6, and which one depends on your plan (dates checked October 10, 2026):

• Free and Go: GPT-6 Luna. It costs nothing on Free, and you can even start without an account.
• Plus, Pro, Business and Enterprise: GPT-6 Sol, the stronger chat model. Plus costs $20 a month.
• Plus: also gets GPT-6 Astra in ChatGPT Work and Codex (limits apply).
• Pro: adds Pro reasoning with GPT-6 Astra in chat.

My advice: start on Free. If you're coming from GPT-4.5 because you want better writing or fewer made-up facts, the jump from Luna to Sol on Plus is the one that actually changes the model, and you can decide after a week of normal use. I walk through every plan, what it costs and which GPT-6 it gets in my ChatGPT review.

GPT-4.5 is a good reminder of how fast this moves: a model OpenAI launched as its biggest chat model in February 2025 was out of the API by summer and out of ChatGPT the next year. If you're reading old articles that recommend it, that's the date to look at first.