OpenAI debuts o3-pro for ChatGPT Pro and Team users and in its API, costing $20/1M input and $80/1M output tokens; Enterprise and Edu will get access next week
OpenAI has launched o3-pro, an AI model that the company claims is its most capable yet. — O3-pro is a version of OpenAI's o3 …
TechCrunchKyle Wiggers
Context & Ripple Effects
OpenAI introduced o3 as its advanced reasoning model alongside a lower-cost o4-mini in April. o3-pro extends that product ladder into a higher-capability option for paid ChatGPT users and API customers, while Enterprise and Edu access follows on a defined rollout path.
ChatGPT Pro and Team users, plus API developers, can choose o3-pro now; Enterprise and Edu customers are queued for access next week.
At $20 per million input tokens and $80 per million output tokens, o3-pro establishes a materially lower published price point than the prior o1-pro offering while preserving a premium output-token charge.
Second-order effects
Developers can re-evaluate which reasoning tasks warrant a premium model, particularly as the standard o3 tier becomes cheaper; model selection becomes a cost-and-performance routing decision rather than a simple upgrade path.
Enterprise AI teams will need to separate high-value, output-heavy workflows from routine tasks, since output pricing makes usage patterns—not just prompt volume—a central budget variable.
Third-order effects
If OpenAI continues releasing capability tiers alongside sharp price changes, inference pricing will increasingly resemble a managed compute market: differentiated by performance, service mode, and workload economics.
The likely durable competition shifts from headline model intelligence alone toward the cost per completed task, with buyers demanding stronger evaluation and AI-spend controls before broad deployment.
The trend: Reasoning models are becoming a tiered commercial service, with vendors using pricing and access segmentation to match expensive inference to narrower, higher-value workloads.
we are going to take a little more time with our open-weights model, i.e. expect it later this summer but not june. our research team did something unexpected and quite amazing and we think it will be very very worth the wait, but needs a bit longer.
In expert evaluations, reviewers consistently prefer OpenAI o3-pro over o3, highlighting its improved performance in key domains—including science, education, programming, data analysis, and writing. Reviewers also rated o3-pro consistently higher for clarity, comprehensiveness, …
o3-pro: it thinks for a *long* time, should be given very very very long and specific instructions, and is very effective on tough problems (better than any we've seen). But of course, even o3 after 15 minutes can't escape overtraining on this modification of the old riddle! [ima…
I have many mind blowing examples of o3-pro outputs, but let me quickly share one. In this instance, I've been working with o3-pro to develop immune system 2.0, a modestly ambitious attempt to completely reengineer our immune system 😆 I first asked o3-pro to identify key [image]
I'd give o3-pro 8.5/10 on BaldurBench. The class build it gave was very strong, fully legal, and patch 8 compliant. Lost a few points for minor/superficial hallucinations and not flagging some optimisation options (eg Ethel's hair). Very very impressive. https://chatgpt.com/... […
Been playing with o3-pro for a bit. It is quite smart. One problem it solved where every other model has failed is making word ladder from SPACE to EARTH. (Probably not contamination: the answer is different than the only online answer, which is for EARTH to SPACE in any case) [i…
o3 pro is very useful though definitely still prone to hallucinations that are “out of character” for its intelligence class (wait what did “we” present to the board?) o4 or o5, which I assume will be better on the hallucination front, will be a more solid foundation for pro. [im…
I asked o3 pro to solve a 10 disk Tower of Hanoi game. Apparently done in 13 mins in a sequence of 682 moves. *Apparently because I can't dload the file it made on my mobile right now. [image]
the price for o1-pro is so astronomically different from o3-pro that I've got to wonder if they just tried to set a high psychological anchor and o1-pro's price was mostly arbitrary [image]
o3-pro scored 87.3% on one of the toughest word puzzle benchmarks. The Extended NYT Connections benchmark takes those viral word puzzles you've probably struggled with and makes them even harder by adding extra words as decoys. Out of 651 enhanced puzzles, o3-pro solved nearly 9
o3-pro sets a new record on the Extended NYT Connections, surpassing o1-pro! 82.5 → 87.3. This benchmark evaluates LLMs using 651 NYT Connections puzzles, enhanced with additional words to increase difficulty. [image]
we stuffed o3 pro to the brim with @raindrop_ai context * planning meeting notes * company goals * user feedback * voice memos about strategy * calendar screenshots and unlike anything else it gave us specific arr goals with timelines + priorities that we are actually using
🚨 JAILBREAK ALERT 🚨 OPENAI: PWNED 🍻 O3-PRO: LIBERATED 🫡 Wowee! Our new fren o3 here is slow as molasses but smart as a whip! Definitely a solid upgrade over previous models, and likely the most capable reasoner we've seen thus far. Refusal mechanisms are strong, which will [image…
OpenAI o3-pro is available in the model picker for Pro and Team users starting today, replacing OpenAI o1-pro. Enterprise and Edu users will get access the week after. As o3-pro uses the same underlying model as o3, full safety details can be found in the o3 system card.
If they didn't do a full Preparedness Framework assessment, e.g. because the evals weren't too different and they didn't consider it a good use of time given other coming launches, they should just say that, I think.
This last sentence seems false? The system card does not appear to have been updated even to incorporate the information in this thread. The whole point of the term system card is that the model isn't the only thing that matters.
If o3-pro were the max capability level, I wouldn't be super concerned about this, and I actually suspect it is the same Preparedness Framework level as o3. The problem is that this is not the last launch, and lax processes/corner-cutting/groupthink get more dangerous each day.