OpenAI debuts GPT-5, calling it ‘major upgrade’ from previous models

Almost immediately after the release of GPT-4 in March 2023, OpenAI began getting questions about when GPT-5 would be available.

But instead of that version, the company shipped scores of new models like o1, o3, 4.1, and 4.5. Advanced users were given the option of several versions of those models, all of which had advantages and disadvantages.

Starting Thursday, OpenAI said there will be only one option: GPT-5. Prompts from the company’s 700 million weekly users will be automatically routed to the appropriate version of the model, balancing cost and latency with capability.

For companies that pay à la carte for OpenAI products, they will still have the option of choosing specific models.

With the new release, OpenAI said it is changing the way it handles its strict content policies. Today, ChatGPT refuses to answer questions that would violate them. But GPT-5 instead will answer with the best response that still fits within its user guidelines.

OpenAI didn’t say whether the model had improved its responses about politics. That has become a hot button issue, with both the left and the right side of the political aisle accusing the chatbot of bias.

Calling GPT-5 an “AI model” is a bit of a misnomer. It really acts more like a system of different models acting like one service for customers.

For most people, the simplified interface will be an improvement. But there will be less transparency now on how the model routes those prompts.

When OpenAI demonstrated the model’s coding ability, it suggested making the same request in several different tabs and then picking the best answer. But OpenAI has said in the past that it struggles to meet demand for its services, and that kind of use case will no doubt put more strain on the company’s massive AI datacenters.

OpenAI will need to find ways to use fewer “tokens” on some requests in order to make room for everyone. Exactly how that works won’t be clear to its consumer customers.

But this is a temporary problem. And, in one way, it’s a good problem to have. As AI infrastructure gets built out, users will have access to unfathomable amounts of compute, and the capabilities will increase just because of the sheer volume of GPUs that can be thrown at the problem.

According to an article in The Information, staff departures and technical hurdles have limited OpenAI’s progress, leading to a GPT-5 release that could be somewhat underwhelming: “The improvements won’t be comparable to the leaps in performance of earlier GPT-branded models, such as the improvements between GPT-3 in 2020 and GPT-4 in 2023, one of the people said. And the slowing performance gains OpenAI has experienced over the past 12 months suggest it may be hard for the company to surge ahead of its biggest rivals, at least in terms of AI capabilities.”