Source: Vertex AI — Standard PayGo and Usage Tiers

How Gemini's shared-capacity pay-as-you-go works: spend-based org-level TPM tiers, 429 semantics and endpoint choice.

Contents

Summary

The page that replaced Dynamic Shared Quota documentation. It explains that Standard PayGo runs on a shared pool with prioritized thresholds set by usage tier.

Key claims

  • Capacity is shared; Standard PayGo gives no dedicated capacity or assured throughput under contention.
  • Each usage tier sets an org-level prioritized TPM threshold per model; traffic above it may still be served but is throttled first. No separate RPM limit.
  • Tiers follow 30-day spend on eligible services: Tier 1 $10–250, Tier 2 $250–2,000, Tier 3 $2,000–50,000, Tier 4 above $50,000.
  • TPM thresholds: Pro/Carbon/Argon 0.5M/1M/2M/10M; Flash/Flash-Lite/Hydrogen 2M/4M/10M/50M (tiers 1–4).
  • A 429 means temporary contention, not a fixed quota. Back off exponentially and smooth spikes within each minute.
  • The global endpoint has the lowest 429 likelihood and single-region endpoints the highest.
  • Tiers don’t apply to preview, image or Omni models. PT commitments count toward tier spend.

Feeds

Google Vertex AI

Suggest changes →