Did you even read my message? Why are you calling me a liar? When I open the link you shared, it says:
The model “o3-mini” is not available.
The link you opened specified a model that isn’t available for your org. We’re using the default model instead.
I am logged in using the exact same email and account as the one I got the alert that it was available for. I shared the email in my post and logged in using that account.
I do not have access through the playground nor the api, despite the email explicitly stating I should have access.
Everyone here is saying we got a message that says we have access but upon checking, we don’t have access. I even spoke to a member of your team who confirmed that the account I contacted them from is supposed to have access, and that this is a bug and they are working on it. Yet days later and nothing is fixed. I want a refund.
OpenAI has totally lost my faith in the past few months and now more than ever. This company is a JOKE. Your team doesn’t even talk to each other, that much is clear by this interaction.
it would be fine if that was actually what was communicated. Instead many of us dropped money into the api to upgrade that we can’t use or get back. That is called deceptive marketing. The AI FAQ clearly says o3-mini api is available to Tier 3 - 5 users, not “select” users.
It looks like you have multiple accounts and only one has access / meets the qualifying tier eligibility. I will DM you the correct organization ID to make sure you log into the correct account, but I’d appreciate it if you keep conversations mature and civil in public settings, for the good of the community.
Would it be possible for you to reach out to me via DM regarding my case as well, please? My account has access only though the playground (verified). Thanks in advance! I’m going to run a new kind of ensemble inference-time reasoning on inductive reasoning models using a new method, and currently trying to run the eval sets. It’s of utter importance to me that I’m able to run these evals so that I’m able to provide the benchmark data on HumanEval, GPQA Diamond, MMLU etc. for this approach. Without access too o3-mini-high, I’ll not be able to provide the data in time and fall behind. I’m a bit unlucky with access via Azure because I filed the o-Model series request form late.
You have OpenAI API access too (Playground mirrors API); I think the issue you’re seeing is with Azure. OpenAI Platform accounts aren’t Azure accounts, they’re two different APIs. Access to one model via OpenAI directly doesn’t necessarily mean you have access with Azure, which is run by Microsoft, not OpenAI. You’re welcome to use o3-mini with the OpenAI API, but — sorry to redirect you — if you’re looking for Azure, you can reach out to them directly or join their forum: https://techcommunity.microsoft.com/category/azure.
I upgraded a few hours ago to tier 3. Previously when I upgraded, I got the effects of the new tier immediately. The general limit page states very clearly and without reservation that tier 3 users would have access to o3mini. That was the reason why I upgraded, and apparently many others as well. If this promise which I consider to be an essential component of OpenAI’s service offer to me and many others is not met, and that was clear to OpenAI, I think there is a legal problem. Under German law this would, to the best of my knowledge, a case of not fulfilling a contractual obligation. Maybe US law is more lenient? Now, I won’t sue them over this - I will just explore the Deepseek model world a bit more, and I will also mention it in an update on social media - but I expect someone will. just my 2cents
Thank you for your reply. I re-checked the situation and dismantled my whole software stack because you said that Playground mirrors API. So I debugged through my whole code only to find that the dotenv module I use in my codebase has an issue reading the .env file on disk and populating it with the right values when running NodeJS version >= 23.
This is a very nasty issue. In my case, the environment had an OPENAI_API_KEY set from a higher scope which would “shadow through” because dotenv would not load and populate the actual values. This is why I would get past the auth layer, but fail with the model error. I had to go and abstract my whole codebase now, and implement my own dotenv loader. Turns out I should have done this earlier. After all, it’s just a few lines of code and you can spare on a 3rd party dependency, security issues and their bugs.
I still can’t believe that this is the reality I live in.
Issue resolved. Wrong key was loaded despite putting it correctly in dotenv file. I bet this is some stuff straight out of fever dreams.
Leaving this here as it might be that more people are running into this. If you are facing the issue I reported (works in Playground, does not in API), do yourself a favor and don’t assume that the dotenv loader code you use and that works fine for the past 15 years in production, still does.
Print the env variable out locally to verify it is using 1:1 the key you expect to be present - like you would do as a junior dev who expects nothing to work. In my case I made sure to copy the right key from the right org, had the right env key name in place. Heck, I’m a professional with 20+ years of experience. I know what I’m doing. And with everything perfectly fitting, it was clear to me, that this is not a user issue – it was.
You would never expect that the upstream lib everybody relies on, would suddenly fail you. Turns out, you must expect everything to fail these days. I now fallback to hoping that at least math and thermodynamics hold ;))
From my experience over the years it always took some time to rollout to all the users. And of course when they say “we roll out to all Tier 3-5 Users” they mean the users that are on tier 3-5 in the moment when they announce it.
Nobody said you can just buy access by adding money.
You haven’t supported OpenAI in the past? Then wait now!
It is not about money. It is about experience with the users because they have no idea how much you are going to use it which makes it hard to scale.
So how about you wait? I’ve waited months for GPT-4 - Everyone here had to do this.
This is not a crowded subway station where you just have to scream and push and you get what you want.
It’s a privilege to get access not a right.
It is by far the best model you can get - it is worth the wait.
Just for reference, I did not have access to the model on Saturday but I am able to call it since Monday. That’s evidence that the roll-out is still going for Tier 3 accounts.
I upgraded to Tier 3 but still don’t have access to o3-mini. I think this issue should be explicitly stated so we don’t all end up complaining after upgrading.
nothing wrong with having to wait. But it’s wrong if they give the impression via their limits per tier page that tier 3 users get access and they don’t. This is hopefully only negligence, because otherwise it would be at least close to fraud.
There are use cases pertaining to batch processing and structured output where I would like to evaluate the answer quality which I cannot really do in the plus account
And as for the rest: a promise is an obligation to at least do your best. I might be a bit old-fashioned I admit. They could have just out on their general page regarding limits a footnote at the o3mini limits that that limit applies only to selected users and might be zero for others and new joiners.
By the way, despite the discussion here over the past few days, they haven’t done this yet, which to me indicates intent, not negligence
Yes, we ran another batch yesterday and added more new eligible developers. And by end of this week, we’ll auto-run more batches more frequently to catch up. Moving forward, there will be some small delay to access o3-mini after reaching Tier 3, but it will no longer be ~days.
Can you provide a status page of these actions so the whole world doesn’t need to check this topic? I assume it is quite a ‘popular’ issue, developers running into and already spreading misinformation on the other fronts of the internet when using 3rd-party software (just judging by GitHub issues). That little misdirection was actually created by your company. So fixing it would probably be in OpenAI’s best interest. ‘A small delay’ is relative.