I am using openai models on Azure AI foundry. This is for an enterprise client.
A few weeks ago I got a notification that computer-use-preview is retiring on 14th of January but there is no mention of any retirement plan anywhere where in the documentations, etc.
I understand this was a preview model, but can I ask if there is a plan for replacing it or is it gone for good.
Regards,
UPDATE: the retirement date has lately changes to April but my question still stands, is there plans to retire this model for good or will it be a non-preview version standalone or part of another model?
If you contrast the absence of computer-use-preview from OpenAI’s deprecations list, you see that Microsoft instead lists EVERY model that is not shut off with a “no earlier than” shutoff date, and also is listing current production models, with even all gpt-4o through gpt-5 getting a “no earlier than” date that is a year out.
So, while the Microsoft dates are not a guaranteed termination of availability or live deployments, a transition to OpenAI services hosted by OpenAI might be your replacement.
Thank you for your response. I am well aware of the url I’m visiting, though I appreciate you pointing it out that this is not Microsoft
Regardless of where I access the models, Microsoft will not be able to offer the model if Openai retires it. Also the model mentioned is a preview model, so it is inevitable that it will by definition retire as soon as a production version is available. Hence my question in this very forum. Microsoft is not any more than a distributor and I wanted to know if there are any roadmap devs can share of the production version of computer-use.
Microsoft has shut off many models that OpenAI has no indications of spinning down, such as gpt-4-turbo and all the gpt-3.5-turbo.
Microsoft has shut down gpt-3.5-turbo-instruct, where OpenAI’s timeline is for another 10 months of service. That is a “deprecation”.
The OpenAI deprecations page is the only indication of an API model shut-down that comes out of the company. From prior performance, you can expect model announcements to give six or twelve months notice, depending on whether there is a logical successor in the family.
You can plan long-term, beyond an AI you can send screenshots to not having direct replacement - does a $450,000 server that uses twelve 3200w power supplies, costing more than the median lifetime earnings of the world population, make any sense running that model shape unless it can displace the economic output of those individuals and more? Racks of aisles of them, would they create billings of equivalent human value, being driven by input/output need on tasks?
Yeah, what you say makes sense. though there are already replacements at least for my usecase that don’t need a $450k servers to run, i.e. Fara 7B + a medium size reasoning model achieves similar results to the computer-use-preview and I can run Fara on a reasonable vm. its just the development cost for us to have something we own and manage is far higher than paying a bit extra for on-demand api and be at merci of Microsoft and Openai.
