ChatGPT 3.5: How to Use It (and What Replaced It in 2026)

You can’t use ChatGPT 3.5 anymore: OpenAI removed GPT-3.5 from ChatGPT in July 2024 and shut down the last gpt-3.5-turbo API models on October 23, 2026. To get what people used it for (free, fast, everyday chat help), open ChatGPT on the free plan, which runs a newer, more capable model, and use it the way the old tutorials describe.

The rest of this guide covers what happened to GPT-3.5, how to use ChatGPT today if you were looking for “3.5” specifically, and what developers with old code should do. All details are as of November 2026, and OpenAI changes its model lineup often.

What happened to ChatGPT 3.5

GPT-3.5 was the model behind the original free version of ChatGPT, which is why “ChatGPT 3.5” became shorthand for “the free ChatGPT.” It was retired in two stages.

Stage 1: removed from the ChatGPT app (July 2024)

On July 18, 2024, OpenAI announced GPT-4o mini, a small, low-cost model. According to Search Engine Journal and TechRadar, it became available to ChatGPT Free, Plus and Team users that day and replaced GPT-3.5, which was no longer a selectable option. Before the change, free users who used up their GPT-4o allowance were dropped back to GPT-3.5. After it, they got GPT-4o mini.

At that point developers could still reach GPT-3.5 through the API.

Stage 2: shut down in the API (2026)

OpenAI’s deprecations page lists October 23, 2026 as the shutdown date for gpt-3.5-turbo-0125 and its gpt-3.5-turbo alias. The GPT-3.5 Turbo model page now says the model has been shut down and is no longer available in the API. Some older GPT-3.5 variants, such as gpt-3.5-turbo-instruct and gpt-3.5-turbo-1106, had an even earlier date, September 28, 2026, according to third-party trackers that mirror OpenAI’s list.

How to use ChatGPT today if you searched for “3.5”

If you wanted ChatGPT 3.5 because it was the free option, the free plan is still the answer. You just get a newer model.

1. Go to chatgpt.com in your browser, or install the official ChatGPT app from your phone’s app store. Check that the publisher is OpenAI.
2. Sign up or log in with an email address or a supported account.
3. Type your question or request in the message box and send it.
4. Read the answer, then reply in the same conversation to refine it (“make it shorter,” “explain step 2 in more detail,” “give me three alternatives”).
5. Start a new chat when you switch to an unrelated topic, so earlier context doesn’t confuse the answer.

Which model you’ll actually get

OpenAI has changed the free-plan default several times since GPT-4o mini, moving to GPT-5-family models. Press reports from 2026 disagree on the exact name of the current free default, and it can differ depending on rollout. Rather than trust any article (including this one), check the model name shown in the ChatGPT interface or OpenAI’s Help Center.

Whatever the name, the current free model can do everything GPT-3.5 could. It’s also a lot more capable. GPT-4o mini alone was described by OpenAI as having a 128K-token context window, far beyond the original GPT-3.5.

Prompting tips that carried over from the GPT-3.5 era

A lot of “ChatGPT 3.5 how to use” videos are really about prompting, and that advice still holds:

• Give context. Say who the answer is for and why: “Explain inflation to a 12-year-old” beats “explain inflation.”
• Ask for a format. Request a table, bullet list, email draft or word count.
• Paste the material. For summaries or rewrites, include the actual text rather than describing it.
• Iterate. Treat the first answer as a draft and correct it with follow-ups.
• Verify facts. Newer models are better, but they can still be confidently wrong. Check names, numbers, quotes and citations before you rely on them.

What no longer applies: tips about switching to GPT-3.5 to save your GPT-4 allowance, workarounds for its short memory, or claims that it is “the free model.”

For developers: migrating off gpt-3.5-turbo

If an app, script or plugin still calls gpt-3.5-turbo, it stopped working on October 23, 2026. To fix it:

1. Search your codebase and config files for gpt-3.5 (including environment variables and no-code tool settings).
2. Pick a current model from OpenAI’s models documentation and deprecations page.
3. Swap the model name, then re-test your prompts. Newer models can respond differently in length, tone and formatting.
4. Check pricing and rate limits for the new model on your OpenAI account before deploying.

A note on the recommended replacement: sources disagree on which model OpenAI recommends in place of GPT-3.5 Turbo. Some copies of the deprecations table point to gpt-5.4-mini, while other reports of the same notice name gpt-5.6-terra. Use whatever OpenAI’s live deprecations page lists when you migrate.

One more catch: according to developers tracking the shutdown, a model can still show up in the GET /v1/models list without actually working. Test real requests instead of relying on the list.

Watch out for “ChatGPT 3.5” lookalikes

Because the search term is still popular, some third-party sites and apps advertise “free ChatGPT 3.5.” OpenAI no longer serves GPT-3.5 in ChatGPT or its API, so these aren’t offering the official model, and you can’t easily tell what they do with your prompts. Stick to chatgpt.com, OpenAI’s official apps, or OpenAI’s API, and don’t enter your OpenAI login details anywhere else.

The bottom line

ChatGPT 3.5 is retired everywhere OpenAI offered it. The free ChatGPT plan has taken over its job with a stronger model, and old prompting advice still mostly applies. If you’re a developer, switch any remaining gpt-3.5-turbo calls to a current model listed in OpenAI’s documentation. OpenAI’s lineup changes frequently, so check the official pages for the latest names.