The latest and largest AI language model from OpenAI, GPT-4.5, was eventually released after weeks of anticipation. On the inside, it was called Orion.
Better writing, information, and a more realistic, less hallucinogenic experience are all provided by this research preview for ChatGPT Pro customers. ChatGPT Pro users will have access to GPT-4.5 first, followed by Plus and Team users the following week, and then Enterprise and Education users.
“GPT-4.5 is ready,” said Sam Altman, CEO of OpenAI, on X. It’s a huge, pricey model. We’ve been expanding a lot and have run out of GPUs, so we wanted to launch it to both Plus and Pro simultaneously.
Also Read: OpenAI Intends to Open Data Centers in India
According to Altman, tens of thousands of GPUs will be added to the Plus tier next week, and hundreds of thousands more will follow shortly. It is anticipated that all of these GPUs will be completely utilized.
CEO Jensen Huang stated during the most recent NVIDIA earnings call that test-time scaling and new reasoning models are driving the company’s growing demand for inference.
Inference-time scaling is used by reasoning models such as Grok 3, DeepSeek R1, and OpenAI. He claimed that reasoning models can use 100 times as much computing power.
GPT4.5 is not a reasoning model because it was trained solely using pretraining, supervised finetuning, and RLHF. Unlike the O series of models, it is an extension of the GPT series.
Also Read: OpenAI adds ChatGPT to WhatsApp
The website states that cached input tokens are offered at a discounted rate of $37.50 per million tokens, while input tokens are priced at $75.00 per million tokens. The price of each million output tokens is $150.00.
“This won’t destroy benchmarks and isn’t a reasoning model. It’s a distinct form of intelligence, and I’ve never experienced the enchantment it possesses. Altman added.
The model provides a tenfold improvement over GPT-4 and is more computationally efficient.
The OpenAI team described the development of GPT models during the livestream, showing how GPT-1 was hardly coherent, how GPT-3.5 became the first model that was helpful, and how GPT-4.5 carried on this trend with little improvements.
Also Read: OpenAI debuts Sora Video-Generating Tool
In the same market, xAI recently released its Grok 3, and Anthropic just released their Claude Sonnet 3.7.
Altman had already revealed the GPT-5 roadmap. In order to eventually produce a more powerful model that might be referred to as artificial general intelligence, or AGI, OpenAI plans to merge its massive language models.