The model list presented for OpenAI doesn’t seem to be updated to match the latest available from OpenAI. How can I refresh and/or reset this list. I’ve tried loading a new OpenAI api key to see if that would trigger a refresh but it hasn’t.
Thanks for any ideas or help
The list includes only handpicked and tested models. The next release of DEVONthink will update the list.
Interesting… I’m not able to get anything in the list of OpenAI models to work… is that expected? (Apologies if you don’t know but based on your answer, it seems you have inside knowledge)
Did you enter your API key in Settings > AI > Chat and add some credit?
Yes, I even created a new API key. But since the new models aren’t listed I wonder if the flow is working at all with openai?
Using claude as a selection (by providing an api key) is working, as is local ollama support.
It seems that only openai interaction is not working.
What new models ?
the new GPT-5 models… these are the ones that the openai API returns (using the same API key that I specified in DEVONthink)
gpt-5-nano
gpt-5
gpt-5-mini-2025-08-07
gpt-5-mini
gpt-5-nano-2025-08-07
gpt-5-chat-latest
gpt-5-2025-08-07
What I mean, is since these don’t show, I wonder if the OpenAI interaction is just not working.
In any case I can’t get anything to work with OpenAI in DEVONthink4 and am just trying to understand what I need to do.
As mentioned above…
The list includes only handpicked and tested models.
So no, the GPT 5 models aren’t going to suddenly appear. DEVONthink isn’t a frontend to AI services. It is controlled and tested access to specific services and models. So no, adding an API key won’t add those new models.
And what problem are you having using the existing models?
PS: As far as @cgrunenberg having “inside knowledge”, he’s our CTO and Lead Developer.
Ok, I understand from @cgrunenberg’s earlier response that you have hard-coded which models are available (as an aside I think that’s an interesting choice, but that would be a separate discussion)….
PS: As far as @cgrunenberg having “inside knowledge”, he’s our CTO and Lead Developer.
Good to know! (I’m new to the forum obviously)
The problem that I have with choosing any OpenAI model is this:
You exceeded your current quota, please check your plan and billing details. For more information on this error, read the docs: https://platform.openai.com/docs/guides/error-codes/api-errors.
I get that reply for any query where I’ve selected an OpenAI model. If I leave the query in place and change to a Claude model or to a locally hosted ollama model, the query works.
I’ve confirmed that the API key I have specified in the settings is valid; as noted I used that same key in a python script to dump the currently available openai model names.
And have you given OpenAI money for API access? It is not the same as e.g., having a ChatGPT Plus account (and they specifically say API access isn’t included in such an account).
Have you read the Getting Started > AI Explained section of the built-in Help and manual?
Ok, I’ve reset everything, restarted, DT4, regenerated a new API key and installed it, and confirmed that I have an account balance and now it’s working.
Don’t know what happened but I’ll assume user-error… anyway it’s working, but I would like to have access to the new models!
Thanks for the responses!
You’re welcome and spend some time reading the section I pointed you to. As I mentioned, DEVONthink isn’t a frontend to your AI provider. It’s handled in a controlled, data-safe, and tested way, with an eye toward privacy and making sure it behaves inside DEVONthink.
Thanks! I have read that link…. but I’ll re-read again.
What I’m hoping for (someday? now?) is to be able to use my huge collection of DEVONthink docs/files in a RAG setup without writing a lot of additional code to extract the docs and put them into a vector store… so any hints on that are welcomed.
Also appreciate the statement on design philosophy.
Thanks again.
The problem with current models is cost, hallucination rate and reasoning error. ChatGPT 5 is an incremental but important improvement on each of those fronts. I infer you’ll build on it soon enough, but appreciate that OpenAI themselves don’t seem to have settled on what the product offering looks like yet.
Incidentally, the implementation of AI in DEVONthink is by far the best approach for legal work I have ever seen (at least for sole practitioners, which a very substantial proportion of lawyers are). That’s an enormous market. There are a lot of people going shopping about now.
Thanks for the feedback and nice comments!
GPT5 is under investigation. In some aspects, it has not been an improvement; in others, perhaps so. Testing and tuning and thinking in here.
Totally agree. (I do expert work so attorneys are my clients - so my needs are very close to yours.)
Try writing some scripts where you can do multiple passes on a long document -first to extra key points and then to format that information in various formats such as a chronological table, bullet points, list of quotes, list of key points and rebuttals… obviously it’s just for idea generation and indexing, not to author your briefs. But it’s incredibly helpful to have next to you as a resource as you do your writing.
I don’t think this question was addressed above. When OpenAI released GPT-5, they removed the other models, and after complaints, reinstated 4o (but IIRC, only for paying customers). When this happens and the models listed in DTP aren’t updated yet, what happens? Is access to those models lost? Or is there some legacy access granted for software that uses these models?
How would you imagine that to be possible? If the provider removes or shuts down the API, then that’s that. Their paying customers might tell them to reinstate the old model, but DT isn’t in that group.
They were only (temporarily) removed from the ChatGPT.app but never from the API.
