Simply hours after saying a giant value minimize for its o3 reasoning mannequin, OpenAI made o3-pro, an much more highly effective model, accessible to builders.
o3-pro is “designed to suppose longer and supply essentially the most dependable responses,” and has entry to many extra software program software integrations than its predecessor, making it probably interesting to enterprises and builders looking for excessive ranges of element and accuracy.
Nonetheless, this mannequin will even be slower than what many builders are accustomed to, gaining access to pc instruments that OpenAI claims make the mannequin extra correct.
“As a result of 03-pro has entry to instruments, responses sometimes take longer than o1-pro to finish. We advocate utilizing it for difficult questions the place reliability issues greater than pace, and ready a couple of minutes is definitely worth the tradeoff,” the corporate mentioned in an electronic mail to reporters.
COME SEE OpenAI Head of Product, API Olivier Godement in-person at VB TRANSFORM 2025 in San Francisco June 24-25. REGISTER NOW whereas tickets nonetheless stay.
However how for much longer? We requested OpenAI about how a lot slower o3-pro is than o3 on common to output responses and can replace once we obtain considered one of our personal from the corporate.
On X, Hyerbolic Labs co-founder and CTO Yuchen Jin posted a number of screenshots of his o3-pro utilization exhibiting it took 3 minutes and $80 price of token to answer the phrase, “Hello, I’m Sam Altman.”
Bindu Reddy, CEO of Abacus AI, additionally mentioned o3-pro took 2 minutes to answer “hey there.”
https://twitter.com/bindureddy/standing/1932562799772971295
use OpenAI o3-pro
Builders can entry o3-pro by the OpenAI API in addition to for Professional and Group customers of ChatGPT. The brand new mannequin model replaces o1-pro within the mannequin picker for paying ChatGPT customers.
OpenAI mentioned o3-pro “has entry to instruments that make ChatGPT helpful,” resembling net search, analyzing information, cause about visible inputs, utilizing Python, and personalize responses.
The mode, although, is costly, which can give some enterprise builders pause. Primarily based on OpenAI’s pricing web page, o3-pro prices $20 per enter and $80 for outputs, in comparison with o3 itself, which is now right down to $2 and $8, a tenth of the worth.
A extra complete mannequin
OpenAI launched o3 and o4-mini in April, increasing its “o-series” of fashions that depend on reasoning and may “suppose with photos.” The brand new mannequin, o3-pro, makes use of the identical underlying mannequin as o3.
Evaluations carried out by OpenAI confirmed the o3-pro can typically outperform the bottom mannequin. Skilled reviewers ranked o3-pro larger in domains resembling science, schooling, programming, enterprise, and writing assist. The corporate mentioned o3-pro is simpler, extra complete, and follows directions higher.
Reasoning fashions have grow to be a brand new battleground for mannequin suppliers, with opponents like Google, Anthropic, and xAI, in addition to rivals from China, resembling DeepSeek, popping out with their very own fashions designed to suppose by responses.
At the moment, o3-pro will not be in a position to generate photos, and OpenAI has disabled momentary chats to resolve a technical subject. ChatGPT’s expanded workspace function Canvas can also be not but accessible utilizing o3-pro.
Some early customers declare that o3-pro has been working remarkably, however it’s nonetheless early days, and the excessive value of operating it might deter some builders from experimenting with it.
See some preliminary reactions under:
As Ben Hyak, a former Apple Imaginative and prescient Professional interface designer and co-founder of startup Raindrop AI observability options wrote in a weblog submit about his early entry utilization of o3, “It’s noticeably higher at discerning what it’s setting is; precisely speaking what instruments it has entry to, when to ask questions concerning the exterior world (relatively than pretending it has the data/entry), and selecting the best software for the job.” OpenAI founder and CEO Sam Altman additionally highlighted Hyak’s weblog submit in an X submit.
The launch additionally comes at a time when OpenAI mentioned it has reached three million enterprise customers, with enterprise customers surging 50% since February.
