OpenAI has launched a research preview of its latest AI model, GPT-4.5, granting early access to Pro users and developers worldwide. The company plans to expand availability in the coming weeks to more users.

What's New in GPT-4.5?

GPT-4.5 brings several notable improvements, including:

Additionally, OpenAI states that the new model has a lower tendency to "hallucinate" (generate incorrect information). GPT-4.5 reduces hallucinations to 37.1%, compared to 61.8% for GPT-4o and 44% for the o1 reasoning model.

Expansion to More Users

OpenAI has scheduled GPT-4.5's phased rollout as follows:

Challenges with Scaling

According to OpenAI CEO Sam Altman, GPT-4.5 is a highly advanced and resource-intensive model. He admitted that GPU shortages have delayed full deployment, though tens of thousands of additional GPUs will be added next week to support expansion.

Supported Features

Currently, GPT-4.5 can handle:

However, it does not yet support voice or video capabilities.

With OpenAI's ongoing infrastructure expansion, GPT-4.5 is expected to bring more advanced AI capabilities to a broader user base soon.