Fast Mode Is Not Offered on Every Model

The mistake

Writing a cost-planning script or a model-picker workflow that assumes every model supports Fast mode as a latency-premium option, because most of the roster does — and having it silently fail, or silently fall back to a wrong assumption, on the handful of models that don't publish a Fast row at all.

Why this happens

Fast mode (renamed from Priority processing) is genuinely available on most current models, which makes "every model has a Fast option" an easy pattern to internalize from experience with the ones you've used most. It isn't universal, though — several models, including some of the largest and most capable ones on the roster, simply don't offer it, and their pricing tables have no Fast row to price against at all, not a Fast row priced at parity with Standard.

Why it matters

Code or planning that assumes a Fast price always exists for a given model will either error against a missing entry or, worse, silently substitute a wrong number (Standard's price, or another model's Fast price) if it isn't handling the absence explicitly. Either way, a workload budgeted assuming Fast mode is an option on a model that doesn't offer it will need re-planning once that assumption turns out to be wrong — ideally before the workload ships, not after.

The fix

Check whether Fast mode is actually published for the specific model in play before assuming it, rather than generalizing from the majority of the roster. Where Fast isn't offered, Standard is the fastest synchronous option available for that model — there's no faster paid tier to fall back to, which is itself worth knowing when comparing latency-sensitive options across models that don't all offer the same tier lineup.

The models missing Fast mode aren't a random scattering, either — they tend to cluster among the largest and the smallest ends of the roster rather than the middle, which is worth knowing if your planning process reasons about tiers "by model size" as a shortcut instead of checking each model's own published table directly.

See also

The token-cost estimator will tell you plainly when a service tier isn't offered for the model you've selected, rather than silently substituting a different number.