See what your Gemini usage costs, per model

Cost anatomy

What actually drives your Vertex AI bill

Per-token, priced by model tier

Vertex AI / Gemini bills per input and output token, separately, and the rate varies widely by model tier. The cheapest and most capable Gemini models can differ several-fold on the same million tokens.

Long-context surcharge

Requests above certain context-length thresholds (e.g. 200K input tokens on some models) bill at a surcharge over the standard token rate. A workload that grows its prompts over time can cross that line without anyone noticing.

Batch vs. online serving

Batch/Flex prediction is priced at roughly half the cost of standard online serving for the same model. A non-realtime workload, like bulk classification or nightly summarization, run online instead of batch leaves that discount on the table.

Context caching, its own line item

Reusing a long system prompt or knowledge-base context via context caching bills cached input tokens far below the standard input rate, plus a separate cache-storage charge. Skip it, and a repeated-context workload pays full input price on every call.

Same billing export as the rest of GCP

Vertex spend surfaces through the same GCP Billing Export in BigQuery as every other Google Cloud service, so it's easy for it to blend into an undifferentiated "GCP" total unless it's split out by service.