Reasoning fashions, which dedicate time to completely considering by way of prompts, are on the entrance of the generative AI pack. OpenAI introduced on March 20 o1-pro, a heavy-duty model of considered one of its reasoning fashions.

Builders on any of OpenAI’s paid tiers can discover o1-pro within the OpenAI API beginning right this moment. These with larger utilization tiers can have larger request and batch queue limits.

The value displays better compute energy: o1-pro prices $150 for 1 million enter tokens and $600 for 1 million output tokens. Examine that to $15.00 for 1 million tokens and $60 per 1 million output tokens when utilizing the bottom mannequin of OpenAI o1.

See: OpenAI Recap: o3 Mannequin Wraps 12 Days of Bulletins

What capabilities does $600 per a million tokens unlock?

The value level makes OpenAI o1-pro one of many priciest AI fashions. It takes the crown from API entry to the Silicon Valley darling’s personal GPT-4.5. (GPT-4.5 prices $75 per million enter tokens and $150 per million output tokens.)

What do you get for that money? OpenAI might be hoping for o1-pro to herald a deep-pocketed handful of researchers, engineers and different professionals utilizing generative AI for science, drugs, or expertise. Whereas o1-pro is optimized for reasoning duties, OpenAI affords separate fashions for transcription or moderation. A smaller, cheaper mannequin can in all probability meet customers’ wants for these.

For the worth of o1-pro you get a 200,000 context window and 100,000 max output tokens.

OpenAI mentioned o1-pro can interpret knowledge from machine imaginative and prescient, with textual content and picture enter. Nevertheless, it solely produces textual content output. Within the API it additionally helps:

• Perform calling.
• Structured Outputs, which make sure the mannequin’s response will conform to the developer’s JSON Schema.
• Integration with the Responses API, which can be utilized to create AI brokers that may work together with the net in varied methods, together with operating searches.
• Integration with the Batch API, which permits for asynchronous requests, offering decrease prices and better charge limits for jobs with a 24-hour turnaround time.

OpenAI has not publicly specified the precise data cutoff date for o1-pro, however earlier fashions had real-world data of as much as late 2023.

AI firms compete to dominate the “reasoning mannequin” panorama

OpenAI launched o1 as “Strawberry” in September 2024. It competes with DeepSeek’s R1, Anthropic’s Claude Sonnet 3.7, Grok 3, Google’s Gemini 2.0, and different reasoning fashions. Meta is engaged on “principle of thoughts reasoning,” however that’s a barely totally different matter: Meta says it’s an adversarial challenge for evaluating superior AI fashions.