The tiers you will see
Standard is the default: real-time processing on shared capacity, best effort. Around it, vendors sell two directions.
- Cheaper and slower. OpenAI's flex processing prices tokens at Batch API rates, with prompt-caching discounts on top, in exchange for slower responses and occasional resource unavailability. Google's pricing page lists flex rates that match its batch rates. The batch API itself is the asynchronous version of the same trade.
- Faster and pricier. OpenAI renamed priority processing to Fast mode on July 30, 2026; it charges a per-token premium over standard and promises up to 2.5× faster output on supported models. Google lists priority rates above standard for every model that offers them. Anthropic runs a fast mode for its Opus models as a research preview at premium pricing.
In our indexPriority, fast and flex rates in our index
Counted at build from our price index: published priority (or fast) and flex input rates against the standard input rate among the 167 current public models we price, from 33 vendors. Each rate comes from the vendor's own price page, read between 2026-07-11 and 2026-10-09. How we collect prices.
How a tier is selected
At OpenAI you set service_tier on the request — flex for flex, fast or priority for Fast mode — or change the project's default tier in settings. The response echoes the tier that actually served the request, which matters because it is not always the one you asked for: if Fast-mode traffic ramps too quickly, OpenAI may serve some requests at standard speed and charge standard rates, returning service_tier: "default". As a rule of thumb it suggests growing by no more than 50% every 15 minutes once traffic passes a million input tokens per minute.
Flex can fail outright: OpenAI returns a 429 resource-unavailable error when it lacks capacity, and does not charge for that request. Its suggested fallbacks are retrying with exponential backoff, or retrying on standard processing when finishing matters more than the discount.
Anthropic's Priority Tier
Anthropic's Priority Tier was a capacity commitment rather than a per-request choice: a fixed number of input and output tokens per minute on a specific model, for one, three, six or twelve months, with a 99.5% uptime target and automatic fallback to standard beyond the committed capacity. Anthropic's documentation now states that Priority Tier commitments are no longer available for purchase; organizations with an existing commitment keep it until the contract ends. Its service_tier parameter accepts auto and standard_only for those customers.
Choosing a tier
Put each workload on the cheapest tier its users can tolerate. Evaluations, enrichment and backfills belong on flex or batch. User-facing chat belongs on standard, and on a fast tier only where latency demonstrably costs you users or revenue — the premium applies to every token, including long prompts the user never sees. Because OpenAI states that Fast mode and standard share the same rate limit for a model, moving traffic to Fast mode buys speed, not headroom.