Google Vertex AI & Gemini Enterprise Pricing: Token-Based Cost Models

Why Google's AI Pricing Is Genuinely Complex

Google's enterprise AI stack spans multiple products, multiple pricing models, and multiple delivery mechanisms — and the pricing changes materially depending on whether you access AI capabilities through Vertex AI directly, through Gemini for Google Workspace, or through the Gemini API. Unlike Microsoft Copilot, which is priced as a single per-user add-on, Google's AI portfolio requires enterprises to build a model of their actual usage pattern before any meaningful cost estimate is possible. Our Google Cloud advisory team works with enterprises to build that model as the first step in any GCP AI negotiation. For the broader GCP commercial framework, the GCP partner channel guide covers how AI commitments can be structured into a Private Pricing Agreement.

Vertex AI: The Foundation Model Access Layer

Vertex AI is Google's managed machine learning platform — the primary interface for accessing Gemini models, fine-tuning, batch prediction, and agent deployment in an enterprise context. Pricing on Vertex AI is primarily token-based for generative AI workloads, with rates varying significantly by model and modality.

For Gemini 1.5 Pro (Google's flagship enterprise model as of 2026), pricing is approximately $3.50 per 1 million input tokens and $10.50 per 1 million output tokens for standard context windows (up to 128K tokens). Long-context requests (above 128K tokens) are priced at a premium — approximately $7.00 per 1 million input tokens. For organisations running high-volume inference workloads, these rates compound quickly: a workload processing 100 million tokens per day generates approximately $350 in daily input token costs at standard Gemini 1.5 Pro rates, or approximately $128,000 per year in input costs alone before output token charges.

Gemini for Google Workspace: Per-User Add-On Pricing

Gemini for Google Workspace delivers AI capabilities within Gmail, Docs, Sheets, Meet, and Google Drive. It is priced as a per-user per-month add-on, distinct from Vertex AI's token-based model. Enterprise pricing for Gemini for Workspace (formerly Duet AI) is approximately $24 per user per month at list pricing for the Business tier, with Enterprise tier pricing available via negotiation.

The commercial significance: for Microsoft-heavy organisations comparing Copilot for M365 ($30/user/month) against Gemini for Workspace ($24/user/month), the per-user price difference represents approximately $72 per user per year — or $720,000 annually for a 10,000-user organisation. However, this comparison only makes sense if your organisation has already resolved the Workspace vs M365 platform question, which requires a broader TCO analysis beyond AI pricing alone.

Our Google Cloud advisory team builds the token volume models and use-case cost estimates that Google's sales team won't build independently — then negotiates committed pricing and PPA terms that reflect your actual AI deployment footprint.

Grounding, Agent Builder, and Additional Cost Layers

Enterprise AI applications on Vertex AI typically incur costs beyond the base model token charges. Grounding with Google Search — which enables AI responses to reference real-time web content — is charged per grounding request at approximately $35 per 1,000 requests. Vertex AI Agent Builder (formerly Dialogflow CX and CCAI) has its own pricing model based on sessions, requests, and storage. These cost layers add material expense for organisations deploying agent-based applications or knowledge retrieval augmented generation (RAG) workloads.

Fine-tuning costs on Vertex AI add a further dimension: custom model training is charged per training hour (approximately $3 to $8 per GPU hour depending on the hardware tier), plus ongoing serving costs for the fine-tuned model endpoint. For organisations building multiple domain-specific fine-tuned models, serving cost management — including auto-scaling endpoint policies and batch prediction routing — becomes essential to avoid runaway spend.

Negotiating Committed Pricing for AI Workloads

Unlike GCP compute where CUD structures are well-established, AI pricing negotiation on Vertex AI is still maturing as a practice. Spend-based committed use discounts are available for qualifying enterprise customers — typically requiring a minimum 12-month commitment of $1 million or more in Vertex AI spend. These commitments can produce 20% to 35% discounts off standard token pricing for high-volume inference workloads.

For organisations building the business case for Vertex AI investment, the commercial comparison with BigQuery data infrastructure costs and the broader GCP committed use framework in our hyperscaler commitment comparison provides the full cost context. To build a predictable Vertex AI cost model and structure committed pricing within your GCP agreement, book a confidential advisory call with our Google Cloud advisory team.

AI Cost Models Built on Assumptions Are Budget Disasters Waiting to Happen

Token pricing, grounding charges, fine-tuning costs, and serving expenses combine into a cost structure that surprises almost every enterprise deploying Vertex AI without independent modelling. Our team builds the model before the bill arrives — modelling actual token consumption, grounding frequency, fine-tuning cycles, and agent request volumes based on your documented use cases.

Building an AI cost model for Vertex AI or Gemini? Get independent advice before you commit. Download our Google Cloud AI negotiation guide to understand the full commercial framework before your next GCP renewal discussion.