Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image releases are available:

• 20260712-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed broken cupy installation.
• M144 (workbench-instances - Debian 11)
  • Installed latest packages from upstream dependencies.
  • Fixed a race condition that could cause JupyterLab to be unreachable (HTTP 524) on GPU instances.

Feature

Secure Boot is compatible with GPUs

You can now enable Secure Boot on Agent Platform Workbench instances that have a GPU attached. Secure Boot with GPUs is supported on the workbench-instances-2603 VM image and the workbench-container-2606 custom container, which include a Secure Boot-signed NVIDIA GPU driver so the driver loads under Secure Boot. For more information, see Create an instance.

Colab Enterprise

Security

A Missing Authorization vulnerability was discovered in repositories in BigQuery, Dataform, and Colab Enterprise. An authenticated attacker could potentially escalate permissions and perform cross-tenant repository takeover. For more information, see the GCP-2026-047 security bulletin.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260701-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed a race condition that could cause JupyterLab to be unreachable (HTTP 524) on GPU instances.
  • Fixed an issue where long-running requests (for example, streaming or long-poll connections) could be terminated after about 60 seconds.

Feature

Python 3.12 base containers for Agent Platform Workbench custom containers

You can build custom containers for Agent Platform Workbench instances using Python 3.12 base containers, in addition to the default Python 3.10 base containers. The Python 3.12 standard and slim base containers are available at the following URIs:

• us-docker.pkg.dev/workbench-images/gcr.io/workbench-container-2606:latest
• us-docker.pkg.dev/workbench-images/gcr.io/workbench-container-slim-2606:latest

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260628-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Installed TensorFlow and PyTorch packages to the default kernel.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260624-1604-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed an issue where notebook kernels could become unavailable if the Dataproc plugin failed to load.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image releases are available:

• 20260622-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed a duplicate "Python 3 (ipykernel)" kernel appearing in the launcher.
• M143 (workbench-instances - Debian 11)
  • Installed latest packages from upstream dependencies.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260531-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• M142 (workbench-instances - Debian 11)
  • Installed latest packages from upstream dependencies.
  • Fixed pip availability in the base environment.
  • Hardened JupyterLab authentication configuration.

Feature

Data Science Agent

Generally available: Use the Data Science Agent to automate exploratory data analysis, perform machine learning tasks, and deliver insights from within a Colab Enterprise notebook. To get started, see Use the Data Science Agent.

Generative AI on Vertex AI

Deprecated

Vertex AI Extensions deprecation

Vertex AI Extensions is deprecated and will be shut down after November 26, 2026. We recommend migrating to Agent Platform to avoid service disruption.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260524-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed a Python 3.12 compatibility issue in the JupyterLab server configuration.
  • Hardened JupyterLab authentication configuration.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260510-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260503-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Improved metadata server resolution reliability on Debian 12 by using the recommended metadata.google.internal name.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image release is available:

• 20260426-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.

Feature

Agent Platform Workbench image release

The following Agent Platform Workbench instances image releases are available:

• 20260419-2130-rc0 (workbench-instances-2603 - Debian 12)
  • Installed latest packages from upstream dependencies.
  • Fixed intermittent connectivity failures caused by proxy TLS certificate rotation.
  • The post-startup script now re-runs after an instance auto-upgrade.
• M141 (workbench-instances - Debian 11)
  • Installed latest packages from upstream dependencies.
  • Fixed intermittent connectivity failures caused by proxy TLS certificate rotation.
  • The post-startup script now re-runs after an instance auto-upgrade.

Feature

RAG Cross Corpus Retrieval

RAG Cross Corpus Retrieval is available in public preview. This feature allows you to retrieve relevant contexts or generate answers from multiple RAG corpora simultaneously using the AsyncRetrieveContexts and AskContexts APIs.

For more information, see RAG Cross Corpus Retrieval.

Feature

Anthropic's Claude Opus 4.7

Claude Opus 4.7 is available in Model Garden.

Feature

Visualization cells

Generally available: You can use visualization cells to generate interactive and editable visualizations from within a Colab Enterprise notebook. You can configure the chart type, aggregation, colors, labels, and other aspects of the visualization to help you explore data and discover insights. For more information, see Use visualization cells.

Feature

SQL cells

Generally available: You can use SQL cells to write, edit, and run SQL queries directly from your Colab Enterprise notebooks. For more information, see Use SQL cells.

Feature

Metadata search for RAG Engine

Use schema-based metadata search in Vertex AI RAG Engine. You can define a metadata schema for a corpus, attach metadata to files within that corpus, and use this metadata to filter contexts during retrieval. For more information, see Filter with metadata search.

Feature

Gemma 4 26B A4B IT is available as an experimental launch in Model Garden. This is an open model built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input and generating text output. Gemma 4 26B A4B IT is available as a managed API in Model Garden. To learn more, see Gemma 4 26B A4B IT.

Feature

Vertex AI RAG Engine Serverless mode

Vertex AI RAG Engine Serverless mode is now available in public preview. Serverless mode provides a fully managed database for storing RAG resources that abstracts away database provisioning and scaling. You can seamlessly switch between Serverless mode and Spanner mode, which provides dedicated, isolated database instances.

For more information, see the following:

• Deployment modes in Vertex AI RAG Engine
• Serverless mode
• Managing Spanner mode
• Switching between modes

Feature

Veo 3.1 Lite

Veo 3.1 Lite is available in public preview. This release is our most cost-efficient Veo on Vertex AI model.

For more information, see 3.1 Lite Generate

Announcement

Gemini 2.5 model retirement dates updated

The retirement dates for Gemini 2.5 Pro, Gemini 2.5 Flash-Lite, and Gemini 2.5 Flash have been updated to October 16, 2026. For more information, see Model versions and lifecycle.

Feature

M140 release

The M140 release of Vertex AI Workbench instances includes the following:

• Gemini CLI Update: Minor updates to Gemini CLI's dependencies.

Feature

Workbench Image Release 26.03

Released the new 26.03 image version under the following image family workbench-instances-2603. This major update introduces:

• OS Upgrade: Migration to Debian 12 (Bookworm) for improved security and performance.
• Python Upgrade: Python 3.12 as the default version.
• New Versioning Scheme: Transitioned to a date-based versioning format (e.g., v26.03) to provide better historical context for image updates. For more information on new versioning schema see the latest documentation
• Micromamba CLI: Installed by default
• JupyterLab Environments: Supports only JupyterLab 4.

Feature

Lyria 3

Lyria is available in public preview. You can use lyria-3-pro-preview to generate 184 seconds of audio, or lyria-3-clip-preview to generate 30 seconds of audio.

For more information, see the following:

• Lyria 3 Pro Preview

• Lyria 3 Clip Preview

Deprecated

Imagen generation GA endpoints deprecation

The following table describes image generation endpoints that are deprecated and their replacements. We recommend updating your model endpoints before June 30, 2026, to avoid service disruption.

Deprecated

Video generation GA endpoints deprecation

The following table describes video generation endpoints that are deprecated and their replacements. We recommend updating your model endpoints before June 30, 2026, to avoid service disruption.

Deprecated

Vertex Explainable AI is deprecated. For details, see Vertex AI deprecations.

Feature

Partner model evaluations

The Gen AI evaluation service supports evaluating partner models, such as Anthropic and Llama models. For more information, see Perform evaluation using the console.

Announcement

Vector Search 2.0 is generally available (GA)

Vector Search 2.0 is generally available (GA).

Vector Search 2.0 is a retrieval engine designed to streamline AI development and serve as the knowledge core for AI applications.

Available features:

• Collections unifying data and vectors.
• Auto-embeddings for populating vector fields automatically.
• Hybrid search and ranking combining vector, full-text, and built-in semantic re-ranking in a single parallel query.
• Private Service Connect (PSC), Private Google Accesss (PGA), and VPC Service Controls support.

For more information, see the Vector Search 2.0 documentation.

Deprecated

Video generation preview endpoints deprecation

The following table describes video generation endpoints that are deprecated and their replacements. We recommend updating your model endpoints before April 2, 2026, to avoid service disruption.

Feature

Gemini 3.1 Flash-Lite

Gemini 3.1 Flash-Lite (gemini-3.1-flash-lite) is available in public preview. This release is our most cost-efficient Gemini model and is optimized for low latency use cases for high-volume, cost-sensitive LLM traffic.

For more information, see Gemini 3.1 Flash-Lite.

Feature

Gemini 3.1 Flash Image

Gemini 3.1 Flash Image (gemini-3.1-flash-image) is available in public preview. This release enables high-quality image generation with improved pricing and latency. We recommend using Gemini 3.1 Flash Image when generating images.

For more information, see Gemini 3.1 Flash Image.

Deprecated

Anthropic's Claude 3 Haiku

Anthropic's Claude 3 Haiku is deprecated as of February 23, 2026 and will be shut down on August 23, 2026. For more information, see Partner model deprecations.