OpenAI today announced o3-pro, its flagship reasoning model that uses more compute to "think harder" and provide consistently better answers. This new model will be replacing o1-pro in ChatGPT since it consistently performs better in math, science, and coding.
To help everyone make the most out of the model, o3-pro supports tool calling. So, based on the user prompt, the model can do a web search, analyze files, reason about visual inputs, use Python, personalize responses using memory, and more. This is a big improvement over o1-pro, which lacked access to tools. But the downside is that o3-pro will take more time to respond. OpenAI recommends users use o3-pro in cases where reliability matters more than speed.
According to OpenAI"s evaluations, users consistently preferred o3-pro"s responses over o3 in key domains like science, education, programming, business, and writing help. o3-pro was also rated consistently higher for clarity, comprehensiveness, instruction-following, and accuracy.
The new o3-pro model is available in ChatGPT"s model picker for Pro and Team. ChatGPT"s Enterprise and Edu users will get access to this latest model next week.
For developers, the o3-pro model comes with a 200,000 context window and is priced at $20 per million input tokens and $80 per million output tokens. OpenAI recommends developers use background mode with o3-pro to prevent timeouts. And the model has a May 31, 2024 knowledge cutoff.
The OpenAI team also noted the following as the current limitations of the o3-pro model:
- At the moment, temporary chats are disabled for o3-pro as we resolve a technical issue.
- Image generation is not supported within o3-pro—please use GPT-4o, OpenAI o3, or OpenAI o4-mini to generate images.
- Canvas is also currently not supported within o3-pro.
As OpenAI continues to refine its models, the balance between speed and reliability will likely remain a key consideration for users choosing the right tool for their needs.