For this text-only configuration:
- Model: mistral/mistral-large-3
- ai: 6.0.272
@ai-sdk/gateway: 3.0.185- generateText
- maxRetries: 0
- providerOptions.gateway.only: ["mistral"]
- maxOutputTokens: at most 8192
- System/user messages only
- No tools, search, fallback models or reasoning override
Local tests establish serialization and local retry behavior, not remote billing guarantees.
Could Vercel staff identify authoritative documentation or confirm:
A. The maximum charged attempts per request, including internal retries, timeouts, partial failures and credential fallback. Which conditions, including credential configuration, are required?
B. Whether maxOutputTokens bounds every billable output class, and what enforced limit covers all billable input, including message framing.
C. Applicable additional charges, minimums, rounding and pricing tiers, and how to calculate a conservative maximum charge before sending.
We are not treating an aggregate budget as a strict per-request cap. If the requested bounds are not offered, please state that explicitly and identify the closest supported control and its remaining limitations.
No billable test or account change is requested.