ChatGPT’s GPT-5 Might Arrive This Summer

OpenAI plans to introduce GPT-5 in mid-2024, according to a report from Business Insider. The next-level upgrade of the AI language model that drives ChatGPT has reportedly been demonstrated to select enterprise clients, according to two anonymous sources close to OpenAI.

One of the CEOs who received this demonstration described the new GPT model as “really good” and “materially better” as it uses company-level cases and data to generate outputs. The CEO also hinted that the model has more hidden features, like launching AI agents made by OpenAI that can do tasks independently.

The following snippet of a transcript from Altman’s March 18 appearance on the Lex Fridman podcast also hints at this big release.

When Fridman asked when GPT-5 is coming out, Altman initially responded that he honestly didn’t know. After further questioning, Altman revealed that OpenAI plans to release many different things in the coming year.

“I think before we talk about a GPT-5-like model called that, or not called that, or a little bit worse or a little bit better than what you’d expect from a GPT-5, I think we have a lot of other important things to release first,” Altman said.

In this discussion with Lex Fridman, Altman does suggest that the company plans to release a big AI model this year, but that’s about it. It could be GPT-5, but it could also just be a big upgrade to the GPT models that are already in use. The upcoming model is expected to remain a Large Language Model (LLM) that can accept text or encoded visual input (aka prompts).

As per the Business Insider report, OpenAI is currently training GPT-5. Once training is complete, the model will undergo further safety checks for potential problems prior to its public launch. The timing of the release may be pushed back based on how long the evaluation takes.

Ars Technica reports that the sources cited might not have accurate information, and the release of GPT-5 might be postponed for reasons other than testing. If GPT-5 is ready for testing, that could mean its main training is likely done, with additional tweaks and improvements expected to come.