OpenAI introduces GPT-4.5, its latest and largest AI language model.

After weeks of anticipation, OpenAI has released its newest AI language model, GPT-4.5, internally known as Orion. Currently in research preview, ChatGPT Pro users are the first to experience its improved writing, knowledge base, and more natural interactions with fewer errors.

OpenAI plans to expand access to Plus and Team users next week, followed by Enterprise and Education subscribers. CEO Sam Altman announced the release on X, noting that the model’s size and cost initially limited its availability to only the Pro tier due to GPU constraints.

Altman assured users that tens of thousands of GPUs would be added next week for Plus subscribers, with hundreds of thousands more planned soon to fully support the model’s demand. NVIDIA’s CEO, Jensen Huang, highlighted the increasing demand for inference capabilities, particularly for new reasoning models like those from OpenAI, Grok, and DeepSeek, which require significant computing power.

GPT-4.5 was trained using established methods and is designed as an extension of the GPT series, not a new type of reasoning model. Input tokens are priced at $75.00 per million, with cached input tokens costing $37.50 per million. Output tokens are priced at $150.00 per million.

Altman emphasized that GPT-4.5 represents a different kind of intelligence, offering a unique user experience. The model is also more computationally efficient, providing a tenfold improvement over GPT-4.

During a live presentation, the OpenAI team described the evolution of GPT models, highlighting the significant advancements from GPT-1 to GPT-4.5. Competing models like Anthropic’s Claude Sonnet 3.7 and xAI’s Grok 3 are also emerging in the AI landscape. Altman has previously outlined plans for GPT-5, indicating OpenAI’s long-term goal of developing more capable AI systems.