On Thursday, OpenAI CEO Sam Altman provided a detailed roadmap update for the upcoming generative AI models—GPT-4.5 and GPT-5—via a post on X. He shared that GPT-4.5, which is internally known as Orion, will be the last model from OpenAI that does not employ chain-of-thought reasoning. Following GPT-4.5, OpenAI will release GPT-5, which will incorporate a wide array of advanced features and improvements over previous models. Altman also conveyed OpenAI’s ambition to simplify its product offerings, acknowledging the growing complexity of its current models. He expressed frustration with the existing model picker interface and indicated that OpenAIs goal is to return to a more intuitive, unified intelligence system that just works for users. This approach is aimed at enhancing the user experience by eliminating the need for users to choose between different models for different tasks.

Regarding GPT-5s access structure, Altman clarified that free users of ChatGPT will have access to GPT-5, but it will be restricted to a standard intelligence setting. This level will come with certain thresholds to prevent misuse and ensure responsible use. On the other hand, Plus subscribers, who pay $20 per month, will be able to unlock a higher intelligence level of GPT-5, offering them more advanced capabilities. Pro subscribers, who pay $200 per month, will have access to the highest tier of GPT-5 intelligence, providing them with the most advanced features and capabilities available. GPT-5 is set to be a major release that will act as an integrated system, combining multiple technologies, including OpenAIs o3 model, which is currently the most capable reasoning LLM. Altman revealed that o3 would not be released as a standalone model, but rather integrated into GPT-5. This integration aims to offer a more powerful and versatile system that can adapt to a wide range of tasks and use cases.

Furthermore, Altman emphasized that OpenAI is moving away from using different models tailored to specific tasks. Instead, the company plans to unify its o-series and GPT-series models, creating systems that can efficiently leverage all of OpenAIs tools. These unified systems will have the ability to determine when long-term reasoning is necessary or when quick, efficient responses are more appropriate. This shift aims to make OpenAI’s AI models not only more adaptable but also more widely applicable across different industries and applications. Altman’s vision is to create an AI that is both powerful and flexible, capable of solving a diverse range of tasks with increased accuracy and efficiency.

Disclaimer: This Image is taken from PCMarg.