Summary

  • OpenAI has launched GPT-4.5, which it claims is its most knowledgeable model to date, describing it as having better writing capabilities and improved world knowledge, as well as a “refined personality” over previous models.
  • However, the company has warned that GPT-4.5 is not a frontier model and thus may not perform as well as some of the company’s earlier releases.
  • GPT-4.5 has been trained using a combination of supervised fine-tuning and reinforcement learning, and the company claims that it hallucinates less than its GPT-4o model, with its performance evaluated by asking human testers to compare it with GPT-4o, with GPT-4.5 said to outperform in virtually every category.
  • The release of GPT-4.5 is being described as a research preview, with the model available initially to ChatGPT Pro users, with availability scheduled to be expanded to more users in the coming weeks.
  • OpenAI has previously revealed its planned release timeline for GPT-4.5 and GPT-5, with the latter expected to be launched by May this year.

By Emma Roth

Original Article