I’ve been testing how GPT-5.5 holds persona skill and it did not disappoint. I’m so excited to work with a new persona skill with GPT-5.5. And it’s overall a more effective model. Well done OpenAI
No sign of gpt-5.5 for me here in the UK, any ideas as to when it is due?
Bill
Hi!
Why are you asking if the Python SDK has been updated? What is it you are trying to do?
I spent some time on the weekend reading through the guidance for the new model. One useful takeaway is that it is worth blocking some time to review the updated prompting guide, then eval, or at least vibe-check, whether low and medium thinking settings are already good enough for the task.
That should help save tokens.
I’ve been getting some great results on medium only.
My assumption is that high or xhigh should provide additional leverage if the work packages are structured differently and built around a better scaffold. In practice, that could mean replacing several medium-sized tasks with one better-scoped task on higher settings.
I think that is the direction I will work toward, mainly to free up more time for other things while Codex is churning away.
Yeah I’m the hand holding type so I tend to give stepwise tasks that have fairly small scope each prompt.
I know. Splitting tasks into smaller chunks is a good habit, built from experience, because it is usually the safest way to make steady progress.
On the other hand, if the progress is real, and 5.2 to 5.4 are pointing in that direction, I am willing to try a different approach.
Also, hand-holding is exhausting, especially when working on several tasks across projects in parallel.
Yes I’ve tried doing this approach on up to three different projects in different terminals simultaneously and it’s too crazy. 
In our codebase, we integrate using Python sdk, but that has not been yet updated to include 5.5. Any idea why?
I use the CLI version. It looks promising but it chew an obscene amount of money in comparison with the 5.4 even i high mode. It is not only it cost double of what 5.4 costs. It is consuming token way faster. And the result is not coming 4 times faster or 5 times faster. I know that this is a non technical opinion but i had to stop its usage until either i learn how to use it in a efficient way (if such thing can be done) or they release a patch for the cli that gets rid of the insane token consumption. Thanks
Models gpt-5.x, as evaluated by ChatGPT itself, although very expensive, are hopeless in strict tasks like legal assistants. I found they hallucinate a lot and aren’t capable of sticking to the strict instructions and to the custom KB. Nor do gpt-4x models do it much better.
What is ChatGPT if not an evolved UI on top of gpt-5.x? Therefore the evaluator is the evaluated? If you can get good results from ChatGPT and consider it an authority, then clearly your API harness could be improved.
I don’t think it is 5.5 but other changes made the prior week. My Plus plan now gets 1.5 hours of work for the 5 hour limit and I get about a total of 7-9 hours of total time for the week, now running out of my weekly credits in less than 2 1/2 days.
I bought an additional 1000 credits to see more than 1/2 disappear within an hour or so.
It’s stopped me dead. I. Cant work on anything. One problem eats up 1/2 of my 5 hours limit and I have to stop and think about what I can start that won’t abort before finishing.
finishing my first tests with 5.5. INSANELY EXPENSIVE. On Plus ($20/month) and what would have been a normal 3 hour work sesh (for 5-hour allotment) used up my allotted tokens in few minutes.
I’m off to ChatGPT to see what I can salvage and migrate to Projects the OG way.
I’m not spending $100+/month for this. Sorry guys.
