ChatGPT's new inference model

O3 Pro is considered by many users to be superior to other chatbots in several areas. However, this model has a major drawback and specific applications.

The o3 pro stands out for its ability to handle complex requests. Image: OpenAI.

OpenAI has launched o3 pro in a Pro package priced at $200/month with Team via API. An upgraded version of o3, which was introduced a few months ago, o3 pro is touted by the company as the most powerful version currently available.

All versions with the added word "pro" are associated with the ability to answer more difficult and longer questions. Unlike typical AI versions, the reasoning model processes problems step-by-step, allowing it to operate more stably and reliably in fields such as physics, mathematics, and programming.

“We recommend using o3-pro for difficult questions where reliability is more important than speed, and waiting a few minutes is a worthwhile trade-off,” the company stated. In shared test reviews, o3-pro achieved superior results compared to the o3 and o1-pro versions.

Commenting on this new model, Ben Hylak, a former Apple employee and co-founder of the AI ​​development company Raindrop, said it's much smarter. He compiled a history of all previous meetings at his company, then asked o3-pro to create a plan.

The results were quite impressive, specific, and clearly analyzed—just as he had always hoped a large-scale language modeling (LLM) could achieve. The plan included target metrics, timelines, priorities, and strict guidance on what to eliminate entirely. “It was so specific and well-founded that I had to rethink the future of my company,” he wrote.

The results obtained from o3 pro (left) are more specific and reliable. Photo: Ben Hylak/X.

O3-pro costs $20 per million tokens invested and $80 per million tokens exported when used via the API. This is due to the AI's ability to memorize and process data. One million tokens invested is equivalent to approximately 750,000 words, which is even longer than the book *War and Peace* , as The Verge compares.

OpenAI states that experts consistently rate o3 pro higher than o3 in every category tested. Reviewers also give o3 pro higher ratings for consistency in several criteria such as clarity, followability, and accuracy, particularly in key areas like science, education , programming, business, and writing support.

On AIME 2024, a test assessing the model's mathematical capabilities, the o3 pro scored higher than even the Gemini 2.5 Pro, Google's top AI. Additionally, the model also surpassed Anthropic's Claude 4 Opus in the GPQA Diamond, a doctoral-level scientific knowledge test.

The o3 pro also integrates tools that allow it to search the web, analyze files, use Python for computation and programming, and personalize responses by leveraging memory. Commenting on this aspect, Ben Hylak noted that the tool clearly demonstrates its ability to recognize its surroundings, knowing when to ask about the outside world (instead of pretending to know), and selecting the right tool for each task.

However, the model's biggest drawback lies in its response time, which is even slower than the o1 pro. YouTuber Bijan Bowen agrees with this. "Although the model's response is quite clear, within just a few descriptive sentences, the response time is quite long," he said. Especially in cases with insufficient external data, the model tends to overthink, Ben Hylak added.

O3-pro also has some other limitations, such as the inability to create images, as well as support for the Canvas feature. The temporary chat feature with this model in ChatGPT is currently disabled while OpenAI fixes a “technical issue”.

However, Hylak argues that this is not a model for user-friendly chat like Claude 3.5 Sonnet or ChatGPT 4o. Nate B. Jones, Head of Product at Rockerbox, advises that the o3 pro should be used for challenging tasks requiring 15-20 minutes of thought.

Source: https://znews.vn/mo-hinh-suy-luan-moi-cua-chatgpt-post1560084.html