Gemini and Vertex AI cost tracking, by model and token type

What CloudQuell reads

Generative AI line items from your Detailed usage cost export in BigQuery, under the Vertex AI and Gemini API services. Google renamed Vertex AI features under Gemini Enterprise Agent Platform in April 2026; its billing catalog still lists them under the Vertex AI service, newer "Agent Platform" SKUs included.

Model

Parsed from the SKU description, so "Gemini 3.1 Flash Lite Global Text Output - Predictions" reads as Gemini 3.1 Flash Lite. Gemini, Claude and other Model Garden models, Imagen, Veo, and embeddings each get a line, with spend by model, a model cost comparison table, and spend over time stacked by model.

Model developer

Google, Anthropic, and the other Model Garden vendors, shown as their own spend breakdown.

Token type

Input (uncached), output, cache read, and cache write, counted only where the SKU bills tokens. Provisioned throughput, context-cache storage, Imagen images, Veo video, and grounding are kept as cost without token counts. Gemini 1.x and PaLM-era usage is billed by characters, so it shows dollars but no token counts.

Serving scope

Global versus regional endpoints where the SKU states it, plus US and EU multi-region for Claude.

Project, region, day

Spend by Google Cloud project, by region, and by day or month, with a Project filter to narrow the whole view.

History

Starts where your BigQuery export starts. Google backfills only the current and previous month when an export is first enabled; if you have exported for a year, you get that year.

What the data looks like in AI Spend

One Vertex AI line item, decoded. The SKU description carries the model, the token type, and the serving scope; CloudQuell keeps all three.

How it connects: via your Google Cloud connection

There is no Vertex AI connector. If CloudQuell can read your billing export, it can read your Gemini and Vertex AI spend.

Connector

Your Google Cloud integration: keyless Workload Identity Federation and two read-only BigQuery roles on your billing export. Nothing extra to install, grant, or key in.

When it appears

A Gemini / Vertex view turns on in AI Spend by itself once your bill carries generative AI usage. The integration syncs daily and re-pulls the current and previous invoice months, absorbing Google’s 24–48 hour restatement window.

Counted once

Gemini and Vertex AI spend stays inside your Google Cloud spend. AI Spend is a read-only lens over those same rows, and its All AI total is never added to cloud totals, so the Overview, budgets, and allocation still reconcile to your Google invoice.

Cost basis

The figures follow the cost settings on your Google Cloud integration, the same as every other GCP line.

Plans

AI Spend, Gemini / Vertex AI included, is on every plan. Free covers up to $10K of monthly cloud spend. Alerts and budgets are on paid plans.

What it unlocks

Gemini and Vertex AI get the same treatment as the rest of your AI bill, without being counted twice.

• All your AI spend, side by side

  Switch AI Spend between All AI, Anthropic, OpenAI, Bedrock, Azure AI Foundry, and Gemini / Vertex, for the sources you have. Under All AI, Claude on Vertex AI and Claude bought direct from Anthropic share one model line.

  Anthropic integration →

• Effective $ per 1M tokens

  A blended rate over token-metered spend only, with a note on how much of your Gemini / Vertex bill it covers. Provisioned throughput, cache storage, images, video, and grounding stay out of the rate. Cache-read share and a projected month-end sit beside it.

• Alerts, budgets, and allocation

  Gemini and Vertex AI spend is Google Cloud spend, so per-service anomaly detection, budgets scoped to a project, service, or label, and alerts to Slack, Microsoft Teams, or email cover it on paid plans. Budgets and anomaly detection work per service, so a budget on the Vertex AI service also counts training and notebooks billed there. Route spend to cost centers with the same label and service rules as the rest of GCP.

  Google Cloud integration →

• From an AI agent

  On paid plans, the MCP tool get_ai_spend returns the same rows with provider "vertex", grouped by model, model developer, token type, serving scope, project, region, service tier, or day. Cloud-billed sources are off by default there, so an agent can’t add them to a cloud total by accident.

  FinOps MCP server →

Common questions

Do I need a separate integration to track Vertex AI and Gemini API costs?

No. Gemini and Vertex AI usage billed to a Google Cloud project is part of your GCP bill, so it arrives through your existing Google Cloud connection: the same keyless, read-only access to your BigQuery Detailed usage cost export. Once that spend appears, a Gemini / Vertex view turns on in AI Spend by itself. There is nothing extra to install, grant, or key in.

Can I see Vertex AI costs broken down by model?

Yes. CloudQuell reads the model from each SKU description, so Gemini, Claude and other Model Garden models, Imagen, Veo, and embeddings each get their own line, split into input, output, cache-read, and cache-write tokens where the SKU bills tokens. The model cost comparison table shows each model’s spend against the prior period, its tokens, and its $ per 1M tokens. Gemini 1.x and PaLM-era usage is billed by characters, so it shows dollars but no token counts.

Does Google Cloud Billing already show Vertex AI cost by model?

Partly. Cloud Billing reports can filter to the Vertex AI or Gemini API service and group by SKU, and SKU descriptions name the model, but one model is spread across many SKUs: input, output, long context, caching, batch, priority. Vertex AI also accepts custom labels on generateContent calls, which Google forwards to Cloud Billing for pay-as-you-go requests and ignores under Provisioned Throughput. CloudQuell rolls those SKUs up into one line per model with tokens and an effective $ per 1M, beside your direct OpenAI, Anthropic, Bedrock, and Azure AI Foundry spend.

Is Gemini / Vertex AI spend double-counted with my Google Cloud total?

No. It is counted once, inside your Google Cloud spend. AI Spend is a read-only lens over those same rows: its All AI figure combines direct and cloud-billed AI spend for analysis and is never added to cloud totals, budgets, or allocation, so every number still reconciles to your Google invoice.

Which Google Cloud AI costs show up in AI Spend?

Generative usage: Gemini, Claude and other Model Garden models, Imagen, Veo, embeddings, grounding, and provisioned throughput, from Vertex AI and from the Gemini API billed to a Google Cloud project, plus Claude bought through Google Cloud Marketplace. Fine-tuning a model counts as that model’s spend, without token counts. Vertex AI training, prediction node-hours, Colab notebooks, Vector Search, and similar services are not model spend and stay in your Google Cloud spend only.

What does Vertex AI cost tracking cost?

AI Spend, Gemini / Vertex AI included, is on the Free plan for up to $10K of monthly cloud spend, with Google Cloud counting as one of three connected integrations. Paid plans are a flat monthly fee: Starter $99/month up to $50K of monthly spend, Growth $199/month up to $200K, with a 14-day trial and no credit card. Never a percentage of your bill.