Vercel applies Token Pass credit automatically across AI Gateway auth methods

Token Pass credit now works with any credential

According to a Vercel changelog entry, usage covered by a Token Pass subscription is now credited automatically when it is served through AI Gateway, no matter which credential the request carries. A standard API key, an OIDC token, or an fx login all qualify, so developers no longer need a dedicated Token Pass API key just to spend the subscription's included usage.

Vercel has kept dedicated Token Pass keys as an option, and they can still be created from the dashboard or the CLI. Their remaining purpose is narrow: scoping a key so it can reach only featured models and providers.

What happens when the credit runs out

Behavior after exhaustion depends on the credential. Once a team's Token Pass allowance is spent for the billing period, requests authenticated with a standard API key or an OIDC token switch over to the team's AI Gateway credit balance, which keeps applications running on paid credits. Requests made with a Token Pass key get no such cushion — they stop until the next billing period begins.

That asymmetry leads to a practical recommendation from Vercel: teams that generated a Token Pass key during initial setup should move to a standard API key or an OIDC token so their traffic can benefit from the fallback path.

Data retention stays consistent on the Token Pass path

Requests handled by the Token Pass provider continue to be Zero Data Retention, an arrangement in which providers do not keep customer request data. Fallback traffic instead follows the team's own retention configuration. Vercel adds that when team-wide ZDR is switched on, fallback requests are routed exclusively to providers that are also ZDR, so the stricter posture survives a fallback event.

Rolling out now

The automatic crediting is available now for Token Pass subscribers on Pro teams, according to the changelog.

Why it matters

AI Gateway sits between applications and many model providers, and billing friction there tends to become engineering friction. Previously, making sure included usage was actually consumed meant tracking which key a service was using, and the wrong credential could mean paying for capacity a team already owned or hitting a hard stop mid-month. Folding the Token Pass entitlement into every authentication method removes that key-management chore, and the fallback to AI Gateway credits turns an abrupt cutoff into graceful degradation.

For teams with compliance constraints, the guarantee that Token Pass traffic stays ZDR — and that fallback respects a team-wide ZDR setting — means cost optimizations do not come at the expense of data handling commitments. The change is narrow in scope, but it removes one more configuration decision a developer has to get right before a single request is routed.