Scale Tier for API Customers

Once you’ve signed an order form, you can add and remove token units through your developer console⁠(opens in a new window).

Billing starts the moment when Scale Tier token units are first allocated, and runs on a calendar month cycle. Invoices will be sent monthly in arrears.

For billing purposes, tokens per minute (TPM) are calculated by averaging the number of tokens used in 15-minute intervals aligned to the top of the hour (e.g. 3:00 to < 3:15, 3:15 to < 3:30, etc). If the total tokens used within a 15-minute period is below your Scale Tier entitlement, they are not billed. For example, if you purchase Scale Tier for GPT-4o with an entitlement of 30,000 input tokens per minute, you can use up to 450,000 input tokens in any 15-minute period without incurring additional charges. Any tokens used beyond this limit are billed at pay-as-you-go (PAYG) rates.

No. Annual commitments lock in discounts on Scale Tier and PAYG. But you’re welcome to use as much or as little of that discounted model as you choose.

No. Annual commitments can be spent on any OpenAI model and any mechanism of delivery (PAYG, Scale Tier, or Reserved Capacity).

Reserved Capacity⁠ for GPT-3.5 and GPT-4 has been superseded with Scale Tier for GPT-4o and GPT-4o mini. If you have an existing reserved capacity contract, the amount of your spend can be directly transferred over to Scale Tier on GPT-4o.

Once Scale Tier is enabled for your account, you can manually adjust your token units.

You can view your TPM aggregated by day. In the future, we will provide you with analytics to view it at a more granular level. For now, please work with your account director to get a custom report.

You can see your current rate limits in your settings page⁠(opens in a new window). When you purchase token units for Scale Tier, your rate limits for that model will automatically increase by the amount of your purchase. When you use the model, requests will first be processed using your faster Scale Tier quota. If you exceed your quota, additional requests will be processed using the regular PAYG service. If you exceed your total rate limit in a minute across Scale Tier + regular PAYG limits, then further requests will be rejected like normal with a 429 error code.

Yes, you can use the ‘service_tier’ parameter within the Chat Completions API. Specifying “auto” will use Scale Tier quota if it’s available. Specifying “default” will use PAYG quota. The response will also include a ‘service_tier’ parameter indicating which service actually processed the request. Values are either scale or default.

We provide a 50% discount on cached input tokens. So, if you send 50,000 TPM in cached input tokens, those tokens only count for 25,000 TPM against your quota. Learn more about Prompt Caching ↗⁠(opens in a new window)

Scale Tier is offered for GPT-4o, GPT-4o mini, and o1 (excluding o1-preview).

Scale Tier supports the same multimodal capabilities available on Pay-as-you-Go. In particular, images can be used as inputs to Scale Tier and are processed with the same fast latency.

At debut, Scale Tier does not support fine-tuned models. We expect to add a Scale Tier SKU for fine-tuned models in time, though the pricing will be different from the base model prices.

You will be credited with the greater of the two SLA amounts for the calendar month of that Scale Tier token unit purchase.

If customers have a use case that qualifies for ZDR, then their Scale Tier usage will adhere to that same retention policy.

Not at this time, though we expect to offer data locality later in 2024, including for Scale Tier.