OpenAI plans to release its next flagship artificial intelligence system, GPT-5, in a matter of months.

Chief Executive Officer Sam Altman disclosed the launch timing today in a series of posts on X. He also shared other details about the ChatGPT developer’s product roadmap.

GPT-5’s debut will be preceded by the release of a less capable large language model, GPT-4.5, in the coming weeks. The latter algorithm is the successor to the company’s general-purpose GPT-4o model. GPT-4o is less adept at tasks such as generating code than OpenAI’s reasoning-optimized LLMs, but it offers significantly lower inference costs.

Altman detailed that GPT-4.5 will be the last OpenAI LLM without a chain of thought, or CoT, mechanism. CoT is a machine learning technique that breaks down a complex task into a series of simpler steps. The method has been shown to improve LLMs’ output quality significantly, particularly when they tackle complex reasoning tasks.

Last year, sources told The Information that GPT-4.5 will provide a “far smaller” improvement in output quality than what its predecessor, GPT-4, delivered at the time of its launch. Additionally, the upcoming model reportedly isn’t “reliably better” than GPT-4 at coding tasks. That might be because OpenAI is focusing its efforts in this area on its reasoning-optimized LLMs, which are specifically optimized for coding and math tasks. 

OpenAI’s newest reasoning LLM, o3, debuted in December. It set a record on Frontier Math, a benchmark that comprises particularly difficult mathematics questions, with a score of 25.2%. The previous best score was under 2%.

In today’s X posts, Altman revealed that OpenAI has scrapped its plans to make o3 available as a standalone product. Instead, the company plans to fold the model into GPT-5, a successor to GPT-4.5 that is expected to roll out in a matter of months. 

“A top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks,” Altman wrote.

OpenAI plans to make GPT-5 available in both the free and paid tiers of ChatGPT. Users with paid accounts will receive access to a version of the LLM with more advanced reasoning capabilities.

Those advanced capabilities might be powered by OpenAI’s test-time compute technology. It enables a model to increase the quality of its output by investing more time and computing resources in processing prompts. OpenAI has already implemented the technology in o3. 

Last month, the company debuted an o3-powered tool called Deep Research that helps users search the web for information. It can generate lengthy reports based on data from hundreds of publicly available sources. Altman today detailed that the free version of ChatGPT will soon enable users to run two Deep Research queries per month, while ChatGPT Plus subscribers will receive the ability to run up to 10 queries. 

Image: OpenAI