How to raise your Azure OpenAI quota

What decides your quota?

Your tier, which Microsoft evaluates per Azure subscription. It looks mainly at your consumption trend across Foundry Models, and if your usage grows until the current tier holds you back, the system moves you up.

Your relationship with Microsoft counts too. Customers on an Enterprise Agreement or MCA-E get assigned higher tiers, and payment history plays a part. The specific offer on your subscription can also change what’s available. Two subscriptions with the same usage can end up in different tiers.

Microsoft doesn’t publish a spend number for each tier.

How do you check your tier?

Through the control plane API. Call the quotaTiers resource on your subscription, and the response gives you currentTierName and your tierUpgradePolicy. Current quota by model is on the Foundry quota page.

Tier	gpt-6-sol Global Standard RPM	TPM
Tier 1	250	250,000
Tier 3	2,500	2,500,000
Tier 6	15,000	15,000,000

Rates as listed on Microsoft’s quotas page, checked November 2, 2026.

The ratio of requests to tokens varies by model, so read the row for your own. The full set is on our AI API rate limits page.

How do you ask for more?

Use the quota request form. You can ask at any time. If Microsoft approves it, your tier stays where it is and the extra quota is added on top.

Or wait. Tier upgrades happen on their own, so if you’re scaling up steadily you may get there without filing anything.

Should you opt out of automatic upgrades?

Only if you use quota as a billing control. Microsoft says that isn’t best practice but knows some teams do it, so you can set tierUpgradePolicy to NoAutoUpgrade and stay in your tier. Microsoft marks this as a preview that may change.

What if you need steady latency?

Microsoft’s usage tiers describe the throughput you can expect with predictable latency. Go past yours during busy periods and response times can rise a lot.

For latency-sensitive or high-volume work, Microsoft suggests Provisioned Throughput Units, which give you dedicated capacity. Tiers don’t apply to provisioned or global batch deployments either, so anything that can wait can run as batch instead.