OpenAI today introduced ChatGPT Pro, a new paid tier of its chatbot that provides access to large language models optimized for reasoning tasks.

The subscription is priced at $200 per month, 10 times more than the consumer-focused ChatGPT Plus plan. The launch of the offering opens a 12-day product announcement series that OpenAI previewed on Wednesday. The company is expected to launch its Sora video generation model during one of the subsequent product reveals. 

ChatGPT Pro provides unlimited access to OpenAI o1, OpenAI’s flagship LLM for reasoning tasks. The subscription also includes o1-mini, a version of the LLM that trades off some response quality for faster response times, and the more general-purpose GPT-4o.

OpenAI first released o1 in preview this past September. Against the backdrop of the ChatGPT Pro plan’s rollout today, the company moved the LLM into general availability with several enhancements.

The latest release of o1 answers prompts faster and more accurately than the preview version. According to TechCrunch, the model is 34% less likely to generate output with “major errors” when responding to “difficult real-world questions.” It has also gained the ability to analyze images uploaded by the user.

ChatGPT Pro customers with more advanced requirements can access a second, more capable edition of o1 dubbed o1 pro mode. According to OpenAI, it uses more hardware resources to process prompts and takes longer to generate responses. The result is a significant increase in output quality.

OpenAI compared o1 pro mode and o1 using AIME 2024, a collection of questions from math competitions. The former model answered 86% of the questions correctly, besting its namesake by 9%. In another benchmark test that involved doctorate-level science questions, o1 pro mode improved upon o1’s score by 3%.

OpenAI found that o1 pro mode performed only 1% better in a test consisting of difficult programming questions. When it comes to simpler coding tasks, however, the model generates 75% fewer errors than o1. OpenAI says those simpler tasks are more reflective of the programming-related prompts that the model is expected to process for users.

Because o1 pro mode takes longer to generate responses, the company has added a progress bar to ChatGPT Pro that tracks wait times. The chatbot also generates a notification when the model completes a task.

OpenAI plans to extend o1’s capabilities over time. It’s working on a function calling capability that will allow the model to perform actions in external applications. OpenAI also plans to equip o1 with the ability to browse the web and process a broad range of file uploads, while ChatGPT Pro will receive “compute-intensive productivity features.”

Image: OpenAI