OpenAI has released an early version of GPT-4.5, the latest significant update to its flagship large language model. This new model is designed to be better at recognizing patterns and drawing connections, with improved hallucination reduction, instruction following, and natural interactions.
The 4.5 model is larger than its predecessors and requires more computing resources to train, making it more expensive to deliver. However, OpenAI CEO Sam Altman claims that GPT-4.5 feels like talking to a thoughtful person, offering a unique kind of intelligence that surpasses traditional reasoning models.
GPT-4.5’s success is due to its “unsupervised learning” approach, where the model’s size and training data are increased to improve its world knowledge and reliability. OpenAI plans to add reasoning improvements on top of this model in the future, which will complement its pre-training phase.
The new model will be available starting today for ChatGPT Pro subscribers ($200/month) and next week for ChatGPT Plus subscribers ($20/month), as well as through OpenAI’s API. The company aims to release GPT-5 models within weeks and months, with the 3o3 reasoning model integrated into its future flagship.
This update represents a significant step forward in AI development and may mark the end of an era for traditional large language models.
Source: https://www.axios.com/2025/02/27/chatgpt-45-model-openai-reasoning