OpenAI, the company behind ChatGPT and its underlying GPT generative AI modesl, is reworking and simplifying its AI models. This should help most people using ChatGPT and similar tools get the best results without understanding the differences between all the models.
Sam Altman, the CEO of OpenAI, announced in a post on Twitter/X that the roadmap for ChatGPT and the company’s other products is changing. He explained, “We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten. We hate the model picker as much as you do and want to return to magic unified intelligence.”
The new plan is to release GPT-4.5 (the ‘Orion’ model that was reported about last year) in the near future, and that model will be the company’s “last non-chain-of-thought model.” It sounds like OpenAI will shift to only developing models with a similar ‘chain-of-thought’ mechanism as GPT o1/o3 and DeepSeek, which breaks down prompts into multiple stages to reduce errors and hallucinations.
OpenAI also plans to release GPT-5 as “a system that integrates a lot of our technology.” It will apparently include the features of the current o3 model, and as a result, o3 will be retired as a standalone model. GPT-5 will support voice, canvas, search, deep research, and other use cases, though it’s not clear if all account tiers will get access to all features.
A “top goal” for OpenAI is to unify all the company’s AI models into “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.” Perhaps, instead of manually picking a model in ChatGPT before entering a prompt, ChatGPT might automatically pick a model based on the complexity of the prompt. The free tier of ChatGPT will get unlimited access to GPT-5 “at the standard intelligence setting,” while Plus and Pro subscribers get an “even higher level of intelligence.”
There aren’t a lot of specific details about the unifying GPT-5 system, but it at least sounds like a solid plan. OpenAI has a lot of AI models with confusing names, and most people don’t understand which models will work better for a specific task. Creating one shared interface that can automatically pick the best model and processes for a task is the ideal solution.
Source: Sam Altman (Twitter/X)
Source link
