Since switching to o3-pro I’ve noticed that each request now takes around 15 minutes to complete. By comparison, the same tasks on o1-pro typically finished in 1–5 minutes. What concerns me is that:
No clear quality gain – The answers I get from o3-pro feel roughly on par with what o1-pro delivered.
Frequent failures – I occasionally receive incomplete responses or outright errors after the long wait.
Effective price hike – Paying the same subscription fee for much slower turnaround feels like a stealth price increase.
I understand that total latency isn’t just model inference time, but a 5- to 10-fold slowdown is hard to justify in day-to-day development work.
Questions for the community (and any OpenAI staff reading):
Is this extreme latency just a launch-period bottleneck, or should we expect o3-pro to remain this slow?
Are there best-practice prompts or workflow tweaks that noticeably reduce o3-pro turnaround time?
Has anyone measured a real-world quality boost that makes the extra wait worthwhile?
Thanks in advance for any insights or work-arounds. The current delay is a major drag on productivity, and I’d love to know whether it’s likely to improve.
I completely agree – at this point, there’s absolutely no incentive to stick with the Pro model. The slowdown with o3-pro has been a huge drag on productivity, and I haven’t seen a meaningful improvement in quality to justify the wait times. If anything, the frequent errors and timeouts make it feel like a downgrade. I’ll give it one more month to see if performance improves, but unless something changes fast, I’ll be canceling. The experience right now just isn’t worth the cost.
I totally agree. When I asked o3 Pro a question, it took more than 24 minutes to answer. Around the 15-minute mark it looked like the model stopped reasoning, but it eventually came back with a reply.
Thanks for your detailed breakdown. That’s a very interesting theory.
I’ve never used the “Deep Research” feature myself, but what you’re saying makes a lot of sense. Funnily enough, I noticed web search capabilities appearing in the o1-pro model just a few days before o3-pro was officially announced.
To be honest, I’m on the fence about how much web search actually improves the quality of the model’s reasoning. Personally, I’d rather get more insightful and creative answers derived from the model’s own “associative thinking” instead of a copy-paste of what’s already being discussed in communities.
On that note, I’m currently testing out Gemini 2.5 Pro, and I think it’s highly likely I won’t be renewing any of my GPT subscriptions
Yes, above all, the Gemini 2.5 Pro is still free for certain queries, especially if you only make a few queries a day. That’s pretty remarkable. And overall, it’s much faster. That’s exactly what really annoys me about the O3 Pro. I was looking forward to a faster, smarter model like O3. And what did we get? An O1 Pro crutch in a new outfit. Once again restricted on all fronts. I’ve now canceled my OpenAI subscriptions and only keep the Plus subscription so I can use Whisper. Because I really like OpenAI’s voice-to-text feature.
