OpenAI Unveils GPT-4.5, Its Largest Conversational AI Model Yet

OpenAI has introduced GPT-4.5, its largest language model to date, designed to offer more natural, intuitive conversations. The company announced the release on its official website.

According to OpenAI, GPT-4.5 enhances writing capabilities, improves world knowledge, and recognizes patterns faster, making it particularly useful for writing and programming.

However, GPT-4.5 is not OpenAI’s most advanced model. According to a leaked document, it is 10 times more efficient than GPT-4 but falls short in performance compared to o1, o3-mini, and Deep Research. That reference has since been removed.

In tests, GPT-4.5 produced fewer hallucinations than GPT-4o and o1, with better factual accuracy. However, unlike o1 and o3-mini, it does not engage in reasoning before responding. OpenAI believes that pretrained models and reasoning-capable models will complement each other in the future.

Running GPT-4.5 is significantly more expensive than GPT-4—input tokens cost 30 times more, while output tokens are 15 times more expensive.

We aligned GPT-4.5 to be a better collaborator, making conversations feel warmer, more intuitive, and emotionally nuanced. To measure this, we asked human testers to evaluate it against GPT-4o, and GPT-4.5 outperformed on basically every category, said OpenAI researcher Raphael Gontijo Lopes.

Currently, GPT-4.5 is available to Pro subscribers ($200/month) and developers via API. Starting the week of March 3, it will roll out to Plus and Team subscribers, followed by Enterprise and Edu users.