I'm using roughly $5,000 in Microsoft for Startups Azure credits for Azure OpenAI and Microsoft Foundry. My main goals are agentic coding in VS Code, complex software development, and scientific or technical research. The subscription has quota for older models, including GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4-32K, GPT-4 Turbo Vision, and GPT-4o, but it shows 0 TPM for the newer models I actually want to use. Requests for about 100K TPM were quickly denied with an "unavailable capacity" message for GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, and GPT-5.4 Mini across Global Standard and EUR Data Zone Standard deployments. The credits are active, so this does not appear to be a billing issue. I have already submitted quota increase requests, tried multiple regions and deployment types, opened Azure support cases, completed automated troubleshooting, requested human escalation, and posted through Microsoft support channels. Has anyone with a startup or sponsored subscription successfully obtained quota for these newer models? Are some regions or deployment types more available? Are sponsored subscriptions restricted compared with paid subscriptions, and would starting with 20–50K TPM instead of 100K improve the chances of approval?
3 Answers
There does not seem to be a reliable escalation path that can create capacity when the service reports it is unavailable. Region and deployment type can make a difference, so it is worth checking several supported regions, but sponsored subscriptions may have some options disabled entirely. Starting with a smaller request could help, though the main issue may be subscription eligibility or regional capacity rather than the requested 100K TPM. Other users have reported the same problem without finding a confirmed workaround.
Sponsored or otherwise non-paid subscriptions often have stricter capacity limits, even when support says there are no formal restrictions. Newer models may be unavailable for quota increases on those subscriptions because capacity is allocated selectively. A practical test is to try a regular pay-as-you-go subscription, although approval still is not guaranteed. Otherwise, use one of the older models that already has quota.
We have seen something similar with an ISV Success subscription. GPT-5 was deployable in Sweden Central using Global Standard, but the actual assigned quota was only about 1K TPM. The deployment screen misleadingly allowed a much larger value, so check the real quota under Management Center → Quota → Standard + Batch and enable “Show all quota.” You may have a small amount available even when a quota request is denied. Trying the minimum quota first can confirm that the model works and establish usage before requesting more.
The deployment setting is not necessarily the quota assigned to the subscription. The quota page is the better source of truth, so I would verify the actual TPM there before assuming the full deployment value is available.