Prices are listed in US Dollars (USD). If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

The costs for Vertex AI remain the same as they are for the legacy AI Platform and AutoML products that Vertex AI supersedes, with the following exceptions:

• Legacy AI Platform Prediction and AutoML Tables predictions supported lower-cost, lower-performance machine types that aren't supported for Vertex AI Inference and AutoML tabular.
• Legacy AI Platform Prediction supported scale-to-zero, which isn't supported for Vertex AI Inference.

Vertex AI also offers more ways to optimize costs, such as the following:

• Optimized TensorFlow runtime.
• Support for co-hosting models.
• No minimum usage duration for Training and Prediction. Instead, usage is charged in 30 second increments.

For Generative AI on Vertex AI pricing information, see Pricing for Generative AI on Vertex AI.

For Vertex AI AutoML models, you pay for three main activities:

• Training the model
• Deploying the model to an endpoint
• Using the model to make predictions

Vertex AI uses predefined machine configurations for Vertex AutoML models, and the hourly rate for these activities reflects the resource usage.

The time required to train your model depends on the size and complexity of your training data. Models must be deployed before they can provide online predictions or online explanations.

You pay for each model deployed to an endpoint, even if no prediction is made. You must undeploy your model to stop incurring further charges. Models that are not deployed or have failed to deploy are not charged.

You pay only for compute hours used; if training fails for any reason other than a user-initiated cancellation, you are not billed for the time. You are charged for training time if you cancel the operation.

Select a model type below for pricing information.

Inference charges for Vertex Explainable AI

Compute associated with Vertex Explainable AI is charged at same rate as inference. However, explanations take longer to process than normal inferences, so heavy usage of Vertex Explainable AI along with auto-scaling could result in more nodes being started, which would increase inference charges.

The tables below provide the approximate price per hour of various training configurations. You can choose a custom configuration of selected machine types. To calculate pricing, sum the costs of the virtual machines you use.

If you use Compute Engine machine types and attach accelerators, the cost of the accelerators is separate. To calculate this cost, multiply the prices in the table of accelerators below by how many machine hours of each type of accelerator you use.

You can use Spot VMs with Vertex AI custom training. Spot VMs are billed according to Compute Engine Spot VMs pricing. There are Vertex AI custom training management fees in addition to your infrastructure usage, captured in the following tables.

You can use Compute Engine reservations with Vertex AI custom training. When using Compute Engine reservations, you're billed according to Compute Engine Pricing, including any applicable committed use discounts (CUDs). There are Vertex AI custom training management fees in addition to your infrastructure usage, captured in the following tables.

• All use is subject to the Vertex AI quota policy.
• You are required to store your data and program files in Google Cloud Storage buckets during the Vertex AI lifecycle. See more about Cloud Storage usage.

You are charged for training your models from the moment when resources are provisioned for a job until the job finishes.

Warning: Your training jobs are limited by the Vertex AI quota policy. If you choose a very powerful processing cluster for your first training jobs, it's likely you will exceed your quota.

Scale tiers for predefined configurations (AI Platform Training)

You can control the type of processing cluster to use when training your model. The simplest way is to choose from one of the predefined configurations called scale tiers. Read more about scale tiers.

Machine types for custom configurations

If you use Vertex AI or select CUSTOM as your scale tier for AI Platform Training, you have control over the number and type of virtual machines to use for the cluster's master, worker and parameter servers. Read more about machine types for Vertex AI and machine types for AI Platform Training.

The cost of training with a custom processing cluster is the sum of all the machines you specify. You are charged for the total time of the job, not for the active processing time of individual machines.

For model-based metrics, charges are applied only for the prediction costs associated with the underlying autorater model. They are billed based on the input tokens that you provide in your evaluation dataset and the autorater output.

Gen AI Evaluation Service is generally available (GA). Pricing change took effect on April 14, 2025.

Computation-based metrics are charged at $0.00003 per 1k characters for input and $0.00009 per 1k characters for output. They are referred to as Automatic Metric in SKU.

Legacy model-based metrics are charged at $0.005 per 1k characters for input and $0.015 per 1k characters for output.

Vertex AI Agent Engine is a set of services for developers to scale agents in production. Services can be used together or a la carte. You only pay for what you use. Today, you pay for the Agent Engine runtime.

Starting February 11, 2026, billing will begin for Code Execution, Sessions, and Memory Bank.

Runtime

Pricing is based on compute (vCPU hours) and memory (GiB hours) resources used by agents that are deployed to the Agent Engine runtime. Billing is rounded to the nearest second of usage. Idle time for an agent is not billed.

Free Tier

To help you get started with the runtime, we offer a monthly free tier.

• vCPU - First 180,000 vCPU-seconds (50 hours) free per month
• RAM - First 360,000 GiB-seconds (100 hours) free per month

Usage based pricing

Once your monthly usage exceeds the free tier, billing begins per the rates below.

Code Execution

Similar to the runtime, you pay for the compute and memory required to run a sandbox. Billing is rounded to the nearest second of usage. Idle time is not billed.

• Compute: $0.0864 per vCPU hour
• Memory: $0.0090 per GiB-hour

Sessions

You pay based on the number of events stored in the session service. We bill for stored session events that include content. This includes the initial user request, model responses, function calls, and function responses. We do not bill for system control events (such as checkpoints) that are stored in the session service.

• $0.25 per 1,000 events stored

Memory Bank

Pay based on the number of memories stored and returned.

• Memory Stored: $0.25 per 1,000 memories stored / month (+ LLM costs to generate memories, paid separately)
• Memory retrieval: $0.50 per 1,000 memories returned; the first 1000 memories returned per month are free

To help you understand the cost of using Agent Engine services, we offer two hypothetical agents: a Lightweight Agent and a Standard Agent. For both scenarios, we make the following assumptions:

• Free tiers: For these calculations, we assume that the runtime and memory bank free tiers have already been used within a month for prior experimentation.
• Runtime Requests per Session: A "full session" or conversation consists of 10 runtime requests.
• Sessions: Each runtime request generates an average of 3 session events.
• Memory Bank:
• Storage: At the end of each full session, 1 memory is extracted and stored.
• Retrieval: We assume an average of 1 memory is returned per runtime request.
• Code Execution is invoked for 30% of all runtime requests.
• Billing Month: All monthly calculations are based on a 30-day month.

Additional notes:

• Service modularity: While the below scenarios show the cost of Agent Engine services used together, you can choose to use services individually. For example, you can use the Agent Engine session and memory bank services without using the Agent Engine runtime.
• Additional model costs: Agents require LLMs to reason and plan. LLM tokens consumed by agents are billed separately and not included in the below scenarios.
• Additional tool costs: Agents require tools to take action. Tools used by agents (e.g. API calls, storage) are billed separately and not included in the below scenarios.

Scenario 1: Lightweight Internal Agent

This scenario represents agents handling low-volume, sporadic traffic.

• Use Case Examples: An IT helpdesk bot for a small company, a personal productivity agent that drafts emails, or a Slack bot that provides answers from documentation.
• Compute and memory required for runtime and code execution: 1 vCPU / 1 GiB RAM.
• Traffic: 0.16 queries per second (10 queries per minute), totaling 432,000 requests per month.
• Average Request Duration: 3 seconds

Scenario 2: Standard Agent

This scenario represents a production agent integrated into a business application, handling consistent user traffic.

• Use Case Examples: A customer service agent on an e-commerce site, a lead qualification bot on a B2B website, or an internal data analysis agent for sales teams.
• Compute: 2 vCPU / 5 GiB RAM
• Traffic: 10 queries per second (600 queries per minute), totaling 25,920,000 requests per month.
• Average Request Duration: 5 seconds

The tables below provide the approximate price per hour of various training configurations. You can choose a custom configuration of selected machine types. To calculate pricing, sum the costs of the virtual machines you use.

If you use Compute Engine machine types and attach accelerators, the cost of the accelerators is separate. To calculate this cost, multiply the prices in the table of accelerators below by how many machine hours of each type of accelerator you use.

• All use is subject to the Vertex AI quota policy.
• You are required to store your data and program files in Google Cloud Storage buckets during the Vertex AI lifecycle. See more about Cloud Storage usage.

You are charged for training your models from the moment when resources are provisioned for a job until the job finishes.

Warning: Your training jobs are limited by the Vertex AI quota policy. If you choose a very powerful processing cluster for your first training jobs, it's likely you will exceed your quota.

The following tables provide the prices of batch prediction, online prediction, and online explanation per node hour. A node hour represents the time a virtual machine spends running your prediction job or waiting in an active state (an endpoint with one or more models deployed) to handle prediction or explanation requests.

You can use Spot VMs with Vertex AI Inference. Spot VMs are billed according to Compute Engine Spot VMs pricing. There are Vertex AI Inference management fees in addition to your infrastructure usage, captured in the following tables.

You can use Compute Engine reservations with Vertex AI Inference. When using Compute Engine reservations, you're billed according to Compute Engine Pricing, including any applicable committed use discounts (CUDs). There are Vertex AI Inference management fees in addition to your infrastructure usage, captured in the following tables.