What is ChatGPT 4? GPT-4 explained simply
ChatGPT 4, usually called GPT-4, was OpenAI's large language model released in 2023. It could read both text and images, and for about two years it powered much of ChatGPT and many business tools built on top of it. Today GPT-4 is no longer available in ChatGPT, and this guide explains what it was, what changed, and why.

What was ChatGPT 4 (GPT-4)?
GPT-4 (Generative Pre-trained Transformer 4) was the fourth main model in OpenAI's GPT series. It followed GPT-3 and GPT-3.5, and was later succeeded by GPT-4o and then GPT-5 (Wikipedia).
A large language model, or LLM, is a system trained on huge amounts of text to predict the next word, which lets it write sentences, answer questions and even generate code. GPT-4 was also a multimodal model, meaning it could accept images as input, not just text (OpenAI). If you want background on OpenAI as a company, see What is OpenAI?.
When was GPT-4 released?
GPT-4 was released on March 14, 2023 (OpenAI, Wikipedia). It became available that day through ChatGPT Plus and the OpenAI API, initially on a waitlist for the API.
A later preview version, gpt-4-turbo-2024-04-09, shipped in April 2024, which was the last major update to the GPT-4 line before it was phased out (Wikipedia).
- Before GPT-4: GPT-3 (introduced in 2020), then GPT-3.5
- After GPT-4: GPT-4o, then GPT-5
| Model | Released | Status today (Nov 2026) |
|---|---|---|
| GPT-3 | 2020 | Long retired |
| GPT-4 | March 2023 | Removed from ChatGPT (2025) and the API (Oct 23, 2026) |
| GPT-4 Turbo | November 2023 | Retired |
| GPT-4o / GPT-5 | Successors to GPT-4 | Superseded by newer releases |
| GPT-6.2 (Astra / Sol) | November 17, 2026 | Live in ChatGPT Work and Codex |
What could GPT-4 actually do?
The biggest change GPT-4 brought was the ability to take images as input alongside text (OpenAI).
- Text and image input: the model could respond based on a photo or diagram, not only written text.
- Fewer made-up facts: on OpenAI's internal factuality tests, GPT-4 scored about 40% higher than GPT-3.5, though it still “hallucinated” (made up facts) sometimes (OpenAI).
- Help with code: a study cited by Wikipedia found GPT-4-generated code had a SQL injection vulnerability (a way attackers can manipulate a database through unsafe input) in 5% of cases, compared with 40% for a 2021 GitHub Copilot model (Wikipedia).
- Steerability via system messages: developers could use a “system message” to shape the model's tone and behavior within limits (OpenAI).
Even so, GPT-4 was not fully reliable: its training data mostly ended around September 2021, and it could still make basic reasoning mistakes (OpenAI).
Hypothetical example: a small online shop owner might have used GPT-4 to draft customer email replies, but would still need to manually check details like prices and delivery dates, since the model could state them confidently and incorrectly.
What was GPT-4 Turbo?
In November 2023, OpenAI announced GPT-4 Turbo. It had two main advantages: a 128K-token context window and significantly lower pricing than the original GPT-4 (Wikipedia).
A context window is simply how much text a model can hold in mind at once, across your messages and any documents you share. A larger context window matters when you are working with long contracts, reports or multi-page documents; see What is a context window in AI? for more detail.
GPT-4 Turbo also combined vision (image) capability with the larger context window, so it could process long text and images in the same request (Wikipedia).
Is GPT-4 still available in ChatGPT today?
Short answer: no. GPT-4 was removed from ChatGPT in 2025, and OpenAI fully removed it from the API on October 23, 2026 (Wikipedia).
ChatGPT now runs on a completely different model family. For example, on November 17, 2026, OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol inside ChatGPT Work and Codex: Astra for complex research and development work, Sol for faster everyday tasks (ChatGPT release notes).
So if an old article or video tells you to “pick GPT-4,” that advice is outdated. Today's ChatGPT users choose from whatever models the app currently lists, and GPT-4 is not one of them. For the bigger picture on how OpenAI's products fit together, see OpenAI vs ChatGPT.
Why does OpenAI retire older models?
OpenAI says directly in its developer documentation that it regularly retires older models (OpenAI developer docs). There are a few practical reasons behind this:
- Newer models are usually more accurate, faster, or cheaper to run.
- Keeping every old version running indefinitely adds server and support costs.
- Engineering and support attention shifts toward the models people actually use.
You can see the same pattern elsewhere in ChatGPT's own release notes: on December 11, 2026, OpenAI is retiring the Advanced and Standard voice modes because almost everyone has already moved to the newer GPT-Live voice model (ChatGPT release notes). GPT-4's retirement followed the same logic.
The practical takeaway: instead of memorizing a model name, focus on the task you need done, whether that's writing, analysis, images or voice. The underlying model will keep changing; your actual need usually will not.
Common mistakes about GPT-4
Searches for “chatgpt 4” are often driven by outdated articles and tutorials, which keeps a few misconceptions alive.
- “GPT-4 was the most powerful model, so I should still use it.” GPT-4 was strong in 2023, but OpenAI has fully removed it from both ChatGPT and the API (Wikipedia); it is no longer an option to select.
- “ChatGPT always uses the same model.” In reality, OpenAI changes and retires models on a regular basis (OpenAI developer docs); today's ChatGPT is not the same product as the 2023 version.
- “I need to know the model name to use ChatGPT well.” For most everyday use, the model name matters far less than how clearly you describe the task; see Is ChatGPT free? for what you get on each plan regardless of model name.
What should you actually do about this?
If you are a regular ChatGPT user, you do not need to track model version numbers. What matters is the result you get, not which model produced it.
- Use whichever model ChatGPT currently offers in its picker; don't search for “GPT-4” by name.
- Judge the tool by the outcome you need, like better writing, faster answers or image understanding, not by a model number.
- If you're comparing ChatGPT with other assistants, see Claude vs ChatGPT or ChatGPT vs Gemini for a practical comparison.
Frequently asked questions
What is ChatGPT 4 (GPT-4)?
GPT-4 was OpenAI's large language model released in 2023 that could accept both text and images as input (OpenAI). It served as the engine behind ChatGPT Plus for about two years.
When was GPT-4 released?
GPT-4 was released on March 14, 2023 (Wikipedia), with a Turbo update following in November 2023 and a final preview build in April 2024.
Is GPT-4 still available in ChatGPT today?
No. GPT-4 was removed from ChatGPT in 2025 and fully removed from the OpenAI API on October 23, 2026 (Wikipedia). ChatGPT now runs on newer model families instead.
Why does OpenAI retire older models?
Because newer models tend to perform better and cost less to run, while keeping old versions active indefinitely adds ongoing infrastructure and support costs; OpenAI states this directly in its developer documentation (OpenAI developer docs).
What is the difference between GPT-4 and GPT-5?
GPT-4's official successors are GPT-4o and later GPT-5 (Wikipedia). Since GPT-4 is no longer available anywhere in ChatGPT or the API, the practical comparison today is moot for everyday users.

