Sam Altman, CEO and co-founder of OpenAI, has just lifted the veil on the future of ChatGPT. In a message published on X, the entrepreneur was keen to clarify OpenAI's "roadmap" for the coming months.
First, the start-up, currently coveted by Elon Musk, will put online GPT-4.5, a new language model. Referred to internally under the code name Orion, the model is an improved version of GPT-4. To converse with it, it will have to be manually selected in the ChatGPT interface.
GPT-5, the model that will unify all the others
This is the last time that OpenAI will offer a new language model in this form. Subsequently, the start-up will do everything possible to "simplify" its offers and the use of ChatGPT. As Sam Altman admits, the use of ChatGPT has become more complex over time. It is no longer enough to ask questions to the AI to use it. You must first choose the version of the most suitable language model and activate the necessary functionalities, such as search. This is far from the ease of use of the first public version of ChatGPT, which dates back to the end of 2022.
After GPT-4.5, OpenAI will therefore return to an interface that will "unify the O series models and the GPT series models". Depending on the user's needs, ChatGPT will automatically use one model or another. You will no longer have to select a model before writing your query.
This is the idea behind GPT-5. Sam Altman describes the model as a "system that integrates a large part of our technology". In short, GPT-5 will replace all existing models. This is the very first version of ChatGPT that will be able to select a model or activate a feature, such as search, completely autonomously, solely thanks to the query.
Among the models incorporated into GPT-5, we find o3. Successor to o1, the linguistic model is capable of thinking and solving complex puzzles, with performances never before achieved for a generative AI. It is mainly intended for scientists. Mirroring the other models, o3 will be activated and used by GPT5 according to the needs of users.
OpenAI's decision probably stems from the stratospheric cost of a query with o3. Designed to replicate the functioning of the human brain, the model costs $1,500 for each request. This is what emerges from the benchmarks of Arc Prize. For OpenAI, it was not profitable to launch this model separately. It is better to slip it into the GPT-5 toolbox. According to Altman, GPT-5 will be offered in the coming months.
GPT5 will be free
The cherry on the cake is that GPT5 will be accessible free of charge to all Internet users. Unsurprisingly, OpenAI will impose a limit on daily messages, as is the case with GPT-4o at the moment. For a long time, OpenAI had been in the habit of reserving its latest models for paying subscribers.
ChatGPT Plus subscribers will nevertheless benefit from a version with "a higher level of intelligence" than the free iteration. Finally, subscribers to the Pro offer, at $200 per month, "will be able to run GPT-5 at an even higher level of intelligence".
With this battery of announcements, OpenAI is probably looking to respond to the increase in competition in the field of AI. The last few weeks have indeed been marked by the arrival of DeepSeek, a very powerful Chinese AI that has blown a wind of panic among the American tech giants. Sam Altman, the boss of OpenAI, had to admit that he finds DeepSeek, whose announced cost does not exceed six million dollars, "impressive".
0 Comments