Huge language models Sam Altman lays out roadmap for OpenAIs long-awaited GPT-5 model GPT-4.5 will arrive in "weeks," then GPT-5 will meld conventional LLMs and reasoning models. Benj Edwards Feb 13, 2025 11:08 am | 8 Sam Altman during the AI Action Summit in Paris, France, on Tuesday, Feb. 11, 2025. Credit: Bloomberg via Getty Images Sam Altman during the AI Action Summit in Paris, France, on Tuesday, Feb. 11, 2025. Credit: Bloomberg via Getty Images Story textSizeSmallStandardLargeWidth *StandardWideLinksStandardOrange* Subscribers only Learn moreOn Wednesday, OpenAI CEO Sam Altman announced a roadmap for how the company plans to release GPT-5, the long-awaited followup to 2023's GPT-4 AI language model that made huge waves in both tech and policy circles around the world. In a reply to a question on X, Altman said GPT-5 would be coming in "months," suggesting a release later in 2025.Initially, Altman explained in a long post on X, the company plans to ship GPT-4.5 (previously known as "Orion" internally) in a matter of "weeks" as OpenAI's last non-simulated reasoning model. Simulated reasoning (SR) models like o3 use a special technique to iteratively process problems posed by users more deeply, but they are slower than conventional large language models (LLMs) like GPT-4o and not ideal for every task.After that, GPT-5 will be a system that brings together features from across OpenAI's current AI model lineup, including conventional AI models, SR models, and specialized models that do tasks like web search and research. "In both ChatGPT and our API, we will release GPT-5 as a system that integrates a lot of our technology, including o3," he wrote. "We will no longer ship o3 as a standalone model."Altman also says that the free tier of ChatGPT users will get "unlimited chat access to GPT-5 at the standard intelligence setting" and that ChatGPT Plus subscribers will be able to "run GPT-5 at a higher level of intelligence, and Pro subscribers will be able to run GPT-5 at an even higher level of intelligence."Altman says these integrated AI models will be able to use many of OpenAI's extended ChatGPT features, he says, including Advanced Voice Mode, Canvas, Search, and Deep Research.Simplifying a confusing array of productsIn the past, Altman has commented about the arbitrary nature of a hypothetical future "GPT-5" label, suggesting that several of the company's AI models released last year could have been christened with that name.In March 2024, for example, Altman said in an interview with Lex Friedman that the company was preparing the launch of a major AI model that year (what likely became GPT-4o or "o1"), but the final name was still up in the air. "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 youd expect from a GPT-5," Altman said to Friedman at the time, "I think we have a lot of other important things to release first."It would be tempting to say that OpenAI may feel a certain threshold of performance has been met to finally label one of its new AI models "GPT-5," but the move apparently comes out of necessity. As mentioned above, GPT-5 will primarily be a branding move that will consolidate many features into one unified interface."We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings," Altman wrote in a post on X. "We realize how complicated our model and product offerings have gotten."Currently, if a user logs in to ChatGPT with a Pro account, they can choose between an array of at least 10 AI models, including GPT-4o, GPT-4o with search, GPT-4o with deep research, GPT-4o with scheduled tasks, o1, o3-mini, o3-mini-high, o1 pro mode, GPT-4o mini, and GPT-4. There's also advanced voice mode. There are two classes of AI models: what you might call "conventional" LLMs (like GPT-4o) and simulated reasoning models (like o1 and o3-mini)."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. Credit: Benj Edwards For an example of the confusion at play, consider that each model offered by OpenAI through this interface has its own strengths and weaknesses with various features attached (such as being able to process or generate images), and it's not immediately clear which one is best for any one type of task. Here's a brief rundown of which is which:GPT-4o: OpenAI's current "standard" and best conventional AI language model.GPT-4o with search: Adds special web search functionality to GPT-4o.GPT-4o with deep research: Uses a special architecture to allow GPT-4o to iteratively perform research on the web and build a report.GPT-4o with scheduled tasks: Allows GPT-4o to perform tasks periodically that include web search and to provide users with updates.o1: OpenAI's "standard" simulated reasoning (SR) model, which is trained to automatically "think through" problems with step-by-step "chain-of-thought" reasoning. Better than 4o for complex logical and mathematical problems but generally worse at writing and some other creative tasks.o3-mini: A smaller, faster version of OpenAI's currently unreleased "o3" SR model. It's the followup of o1 because the name "o2" was unavailable due to trademark issues.o3-mini-high: A version of o3-mini that "thinks things through" a bit more deeply than o3-mini, but it's slower.o1 pro mode: The slowest and currently "best" SR model that OpenAI offers, only available to Pro accounts.GPT-4o mini: A legacy model that was once the free, fast, and cheap version of GPT-4o. Kept around for people who might have prompts that work best with it.GPT-4: OpenAI's last version of the legacy GPT-4 model, first introduced in 2023.Advanced Voice Mode: A special version of GPT-4o designed to interact vocally.Expecting anyone who has not been paying close attention to OpenAI's model development to understand this array of options is probably not realistic, and Altman recognizes that. "We hate the model picker as much as you do and want to return to magic unified intelligence," he wrote in his post.Meanwhile, competitors like DeepSeek, Anthropic, Meta, and Google are hot on OpenAI's heels, each iterating its own AI models at a breakneck pace that never seems to slow down. We'll keep you updated as more news about firm release dates of GPT-4.5 and GPT-5 emerge.Benj EdwardsSenior AI ReporterBenj EdwardsSenior AI Reporter Benj Edwards is Ars Technica's Senior AI Reporter and founder of the site's dedicated AI beat in 2022. He's also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. 8 Comments