What’s the situation with API access these days? I’m looking to launch the trading application, the API waitlist doesn’t seem to be moving. What’s your experience with API access? Is there a way to get it faster or at least at predictable timeframe?
I also looked at Azure’s API equivalent. There is even more paperwork and forms.
Thanks
R
Hi rn1,
There is no method for moving your application through the waitlist. It seems to be based on the time it was first entered into the system, your familiarity with the OpenAI ecosystem and your use case being a unique, scientifically/commercially/educationally exciting one.
I’m afraid it’s extremely boring, enterprise software 
Maybe this in the meantime?? GitHub - 0m364/Bard: Python SDK/API for reverse engineered Google Bard
I have tried google’s model, it’s nowhere near close GPT3, which already has API access. Gpt4 requires much less effort to build application.
The question is about paid access to gpt4 or equivalent commercially available model.
Agreed. That’s just a way I’ve been having my gpt API chat with bard and llama (I’m not using it for work though I’m just playing around as a hobby)
Well, commercially viable enterprise software may be boring on a “wizzbang, look at that level” but it is VERY interesting on a “Oooo look at that bank account level” So they all have their merits. The question then becomes, is the use case unique. There is nothing stopping you from proposing multiple use cases, i.e. applying again with a different idea. Worth a shot?
I mess around with language models as a hobby and do penetration testing as a profession… people see tv and think pentesting is interesting… other than the part that could be done by a qualified locksmith it’s very boring lol… but it does pay the bills… so I see what you are saying
Maybe, the real question is how to subscribe for API to use in a commercial product.
I’m not looking to prove or demo anything or participate in exciting use cases or marketing campaigns or whatever the criteria for approving access. I need to run the app, so I started to look around, and there is nothing commercially available, at least not easily, and not for small firms like mine.
I would like access, I feel commercially it puts people at a disadvantage and is unfair if others can make money off the API access and I and others are not allowed. Feels wrong.
I have wondered the same thing, is it moving? I got plugin access, the request to allow my 120 dollar limit took a long time too, so that is another aspect. Not sure how the commercial product can be controlled that much with a beta service that has such a hot bed of potential right now with early entry key.
I’m pretty sure you have to have ChatGPT plus in order to access the API. that in addition to API access approval. I signed up and paid first before receiving access. Don’t know if it made a difference.
You don’t need ChatGPT Plus to use the API. The only waitlist is for GPT4 API, you can use GPT3.5 on API immediately after signing up for API.
I managed to make it work with 3.5 with a longer and more explicit prompt.
3.5 is more cost effective, too. the prompt itself is rather long, plus data to analyze - and it starting to use lots of tokens. Chatgpt4 may be prohibitively expensive, although it creates much more concise reasoning for selecting specific data points.
Other than 3.5, there is no other API freely available. I’ve sent a bunch of waitlist join requests - and nothing. Well, good for OpenAi 
The waitlist situation is frustrating. What is the primary reason behind the waitlist anyways? Is it limited server resources? If so, I am happy to accept a a rate limit, same as already active on the web frontend. Or are there other reasons?
Yes, these models run on supercomputers, and given all the AI craze, the GPUs that run them are very scarce.
I suppose it would help if OpenAI at least communicated a bit about the state of affairs. Even better if they’d give you your position in the queue, but I can see how that would make it more difficult to change stuff around as they see fit.
Well, I suppose the same is true for the communication question: anything they say about the process can later be used against them (“But last year you said …”).
I’ve had really good results with the 3.5 turbo-16k. I’ve been manually comparing outputs of the models including manually copy/ pasting from the gpt 4 web ui and I guess based on new fine tuning, 3.5 turbo 16k at least presents more of the appearance of gpt 4, and gives a lot more room for the prompt context.
Openai announced GA for gpt-4!
Does anybody can see it in the list of available models? Still missing in mine account.
curl https://api.openai.com/v1/models -H “Authorization: Bearer xxx” | jq -r ‘.data.id’
Have you made a payment for API usage to date? if not you will get access by the end of the month.
I had negligible usage, dimes. The last charge is June, 5