GPT-5 arrives at the beginning of August: what to expect from OpenAI's new model

OpenAI is ramping up the hype around GPT-5. Sam Altman revealed a few days ago that this version would be released « soon », while setting realistic expectations: « it is an experimental model that integrates new research approaches which we will use in our future models ». It won't take much longer to find out: according to information from The Verge, GPT-5 is scheduled for late August.

GPT-5 aims to think better, further, and longer.

In essence, GPT is the engine of ChatGPT: it's thanks to the language model (LLM) that the bot can answer questions. The initial model of ChatGPT dates back to November 2022, and it was GPT-3.5 that allowed OpenAI to introduce generative AI to the general public. GPT has come a long way; since spring, ChatGPT uses GPT-4o for audio and image-related tasks, and GPT-4.1 for code, long tasks, or agents.

GPT-5 is a system that « integrates a lot of our technologies », as Sam Altman described. It indeed incorporates the capabilities of the o3 and o1 reasoning models, which take more time to think but in exchange, provide more accurate and well-supported answers. Merging these different models into one will allow simplifying the overwhelming and confusing ChatGPT offerings — users often struggle to understand the nuances between different models, and even more so to choose the right one.

GPT-5 should also enhance its mathematical, scientific, and programming capabilities to better compete with the latest models from Anthropic (Claude 4) and Google (Gemini 2.5), which are particularly strong in these areas. We are also waiting for the multimodal LLM on the field; GPT-4o already supports text, images, and audio, but GPT-5 could "read" videos.

The model should also be able to process very long and voluminous documents, talking about a x300 improvement compared to GPT-4o.

The integration of different models under the umbrella of GPT-5 could bring LLM and OpenAI closer to general artificial intelligence (AGI). This notion is crucial, as once AGI is achieved, Microsoft would lose its rights to the revenues and future models of the startup. The two partners are currently renegotiating their agreement in pain, especially because OpenAI wants to shift part of its activities towards a for-profit model.

GPT-5 should however not cross this threshold: if it relies on the advanced reasoning capabilities of the o3 series, Sam Altman has clarified that it is still an experimental model. A prototype that has not yet been commercialized has indeed reached the "gold medal" level at the 2025 International Mathematics Olympiad — a symbolic milestone in the quest for AGI — but this performance will not be included in GPT-5 for several months.

ChatGPT takes a big leap forward: with Agent, OpenAI's AI controls your PC and does everything for you

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : The Verge