Google Release Notes

Agent Platform Workbench

v2

Feature

New configuration options for scheduled notebook runs

When you schedule a notebook run from an Agent Platform Workbench instance, you can now configure the following:

• Execution identity. You can choose your own user account instead of a service account. The notebook runs as you, so Application Default Credentials inside the notebook resolve to your identity and the run accesses only the resources you can access. If you have not yet granted consent, the scheduler provides a link for authorizing access.
• Shielded VM options. To meet workload requirements that mandate verified boot integrity, you can enable Shielded VM options, including Secure Boot, on the compute that runs the notebook. For more information, see Modifying Shielded VM options.
• Runtime environment. In addition to the default environment, you can specify a VM image or a custom container, so that scheduled runs use the same dependencies as your interactive environment.

Ensure that scheduler-jupyter-plugin is at version 0.1.8 or later for these features to be accessible. For more information, see Schedule a notebook run.

Feature

Gemini Enterprise: Monday federated data store (GA)

Connecting a Monday data source with Gemini Enterprise using data federation is generally available (GA). You can use the Gemini Enterprise app assistant to search across your Monday boards, items, updates, and docs, and get answers grounded in that content without needing to ingest the data.

For more information, see Set up a Monday data store.

Feature

Gemini Enterprise: Create and manage projects

You can create and manage projects in the Gemini Enterprise web app to build a dedicated knowledge base for your own work or to collaborate with your team. Within a project, you can upload files and chat privately with the assistant to get answers grounded in your project files and web search.

This feature is generally available (GA). To make projects available to users, a Gemini Enterprise administrator must turn on the Enable projects toggle in the Google Cloud console.

For more information, see the following:

• Create and manage projects
• Manage web app features

Feature

Gemini Enterprise: New data stores and support for new actions (Preview)

The following data stores are available in Public Preview in Gemini Enterprise: …

Feature

Gemini Enterprise: Latency and error rate views for agents

To monitor operational telemetry for your agents, use the two new views on the Observability tab:

• Latency: Shows response times for your agents. This view displays p50 (median) and p95 (95th percentile) metrics for Time to First Token (TTFT), Time to First Answer (TTFA), and Time to Last Token (TTLT). TTFT counts the first token of any kind, including the model's thinking, while TTFA counts only the first token of the answer itself. You can also compare latencies by specific agent features, such as web search, media generation, or parametric interactions.
• Error rate: Shows how your agent's requests resolve by tracking request volume and error rates. This view groups results by response class (OK, client errors, server errors, and canceled) and displays the associated client and server error codes.

This feature is generally available (GA). For more information, see Access metrics.

Feature

Gemini Enterprise: General availability of Workflow Builder (formerly Agent Designer)

Workflow Builder (formerly known as Agent Designer) is generally available (GA) in Gemini Enterprise.

Workflow Builder enables users across your organization to build multi-step automated workflows to streamline tasks and connect to enterprise data.

Key capabilities in this release include: …

Firebase AI Logic

• Firebase AI Logic supports the latest Gemini stable model gemini-3.8-flash. You can directly access this model in your mobile or web app via either the Gemini Developer API or the Agent Platform Gemini API (formerly Vertex AI).

  • If you're using the Gemini Developer API, accessing this model does not require the pay-as-you-go Blaze pricing plan.

  • If you're using the Agent Platform Gemini API (formerly Vertex AI) with the legacy "Vertex AI" initialization syntax, make sure to set the location in your request to global.

  • If you're migrating from previous models, note the following for this new model:

    • No support for MINIMAL thinking level: gemini-3.8-flash does not support the MINIMAL thinking level. If you set the thinking level to MINIMAL for this model, the request will fail with a 400 error.

    • Use thinking levels instead of thinking budgets: Thinking budgets have been supported by Gemini 3.x models for backwards-compatibility. However, starting with gemini-3.8-flash, they are strongly discouraged. If you haven't already, migrate to use thinking levels. …

Feature

Gemini 3.8 Flash is generally available

Gemini 3.8 Flash is now generally available (GA) and available for production use.

For more information on 3.8 Flash, see the model page.

Feature

Provisioned Throughput support for Gemini Omni

Provisioned Throughput supports Gemini Omni. To learn more, see supported models.

Fixed

CodeMender updates (v0.6.0)

This release introduces updates to CodeMender:

• Machine-readable metrics: Added the --json flag to cm stats to export aggregate and per-session metrics (CACHE_HIT%, THINK_RATIO%, TOOL_CALLS, DURATION).
• Session drill-down: Added cm stats --session <id> to inspect turn-by-turn token consumption for specific sessions.
• Bug fixes:
  • Improved codebase search reliability by skipping binary archives and non-regular files during traversal.
  • Fixed an issue where cm report import failed on native JSON reports or findings referencing new files.
  • Fixed an issue where preview mode could create empty directories on disk before user confirmation.
  • Prevented erroneous verification verdicts when workspace reset fails.

For more information, see CodeMender documentation.

Feature …

Remote Config

• Starting September 1, 2026, Remote Config offers a usage-based pricing structure based on daily fetch request volume. Every project includes a no-cost tier of up to 100,000 daily fetch requests, with pay-as-you-go pricing on the Blaze plan for usage exceeding this threshold. All features (including Personalization, Rollouts, and A/B Testing integration) remain included across both Spark and Blaze plans at no additional cost. To learn more about pricing plans, grace periods, and usage monitoring, see Pricing for Remote Config.

A/B Testing

• A/B Testing is now integrated directly into Remote Config as a native feature, bringing experiments into the core Remote Config template lifecycle. To learn more, see Create Firebase Remote Config experiments with A/B Testing and the A/B Testing troubleshooting and FAQ guide.

Feature

Anthropic's Claude Fable 5.1 is now available

Claude Fable 5.1 is available in Model Garden.

Change

Embedding SKUs are changing

Memory bank uses the following embedding model SKUs:

• 6E46-5623-C0B6
• C15D-A68F-5C1E
• 2D07-D52C-A93B

For detailed information about individual SKUs, see Google Cloud Platform SKUs.

Feature

Agentic video processing now available in Preview

Gemini 3.5 Pro and later models now support agentic video processing, where the model dynamically navigates video content instead of extracting frames at a fixed rate. This reduces token usage for long-form video and improves answer quality. Set processing: "agentic" (Interactions API) or media_processing: "AGENTIC" (GenerateContent API) on video inputs, or use the model default.

See Video understanding for details.

Change

Pricing updates for agent metering

Session and memory bank compute metering is in effect for the Agent Platform compute SKU. For more information, see Gemini Agent Platform Pricing.

Feature

Gemini Enterprise: Overage controls available for all invoiced Cloud Billing accounts

Configuring overage controls in Gemini Enterprise is available to all projects linked to an invoiced Cloud Billing account. Previously, customers that received an email with the subject line [Billing Update] New Gemini Enterprise overage billing controls launching Aug 17, 2026 couldn't enable overages despite having an invoiced Cloud Billing account. This restriction no longer applies.

For more information, see Overview of overages and spend controls.

Firebase AI Logic

• Firebase AI Logic now supports the us and eu locations. When you use the Agent Platform Gemini API (formerly Vertex AI) (and depending on the model), you can choose where to access the model to meet your compliance, latency, availability, and durability requirements.

  Note that Firebase AI Logic defaults to the following locations:

  • When using the new "Agent Platform" initialization syntax: global
  • When using the legacy "Vertex AI" initialization syntax: us-central1

  Learn more about specifying the location for accessing the model.

Feature

Gemini 3.1 Flash Image and Gemini 3 Pro Image updates

This release introduces feature updates and expanded endpoint availability for Gemini 3.1 Flash Image and Gemini 3 Pro Image:

• Multi-region endpoint support: Gemini 3.1 Flash Image (gemini-3.1-flash-image) is now available on US (us) and EU (eu) multi-region endpoints for model availability, ML processing, Provisioned Throughput, and PayGo Standard.

• 4K image output in GA: Generating 4K resolution images is now Generally Available (GA) for both Gemini 3.1 Flash Image and Gemini 3 Pro Image.

• Video inputs in GA: Image generation from video inputs is now Generally Available (GA) for Gemini 3.1 Flash Image.

For more information, see the following:

• Gemini 3.1 Flash Image
• Gemini 3 Pro Image
• Model locations

Feature

IAM Unified Access Policies are generally available (GA) …

Feature

Gemini Enterprise: Support for federated data stores in Assured Workloads with FedRAMP High compliance

Gemini Enterprise now supports connecting Google Workspace and third-party federated data stores to projects inside Assured Workloads folders, which automatically enforce security and compliance controls on your Google Cloud resources to meet FedRAMP High standards. This feature is generally available (GA). For more information, see

• Connect a Google data source.

• Connect a third-party data source.

Feature

Gemini Enterprise: Gemini 3.5 Flash regional availability in Canada

Gemini 3.5 Flash is available in Canada (ca) with in-region at-rest data residency (DRZ) and machine learning processing (MLP).

For more information, see Locations.

Feature

Gemini Enterprise: Gemini 3.1 Flash image available in US and EU multi-regions

Gemini 3.1 Flash image (Nano Banana 2) for image generation is generally available in the us and eu multi-regions.

For more information, see:

• Manage features on the web app …

Agent Platform Workbench

v2

Change

20260830-2330-rc0 Release

Change

20260831.01_p0 Release

Change

Installed latest packages from upstream dependencies.

Change

Installed latest packages from upstream dependencies.

Fixed

The terminal and file browser are now more responsive when working in a Cloud Storage bucket that is mounted through the file browser. Frequent background checks for non-existent files are now cached instead of repeatedly querying Cloud Storage, which previously could make actions such as listing files or running terminal commands take several seconds.

Fixed

Cloud Storage buckets that you mount through the file browser are now automatically re-mounted after the instance is restarted or is stopped and started. Previously the mounted folder could be left behind as an empty, unusable directory that had to be manually removed and re-mounted.

Change

The obsolete google-cloud-sdk transitional package is no longer installed. The Google Cloud CLI itself is unchanged; it was already provided by the google-cloud-cli package.

Change

20260830-2230-rc0 Release

Change

20260830-2230-rc0 Release

Change

Installed latest packages from upstream dependencies.

Change

Installed latest packages from upstream dependencies.

Fixed …

Feature

Gemini Enterprise: New data stores and support for new actions (Preview)

The following data stores are available in Public Preview in Gemini Enterprise:

• Campfire
• Clay
• CourtListener
• Daloopa
• MSCI

You can search and read data from these data stores using natural language.

Additionally, the following data stores support new actions in Public Preview:

• AirOps: Update knowledge base document metadata.
• Airtable: Create records for a table.
• Zoho Desk: Update event.

Imagen models

• All Imagen models shut down on August 17, 2026 (note that this shutdown date was extended from the previous June 30, 2026).

  To accommodate this shutdown, we've removed all Imagen APIs from the Firebase AI Logic SDKs and documentation.

  This shutdown date is applicable to both the Gemini Developer API (Google AI) and the Agent Platform Gemini API (formerly Vertex AI). As a replacement, you can migrate your apps to use a Gemini Image model (like one of the "Nano Banana" models).

  If you use Imagen models via Firebase AI Logic, you can migrate your apps to use a Gemini Image model to restore service. Models that are shut down are no longer accessible or supported by Google, and a request using a shutdown model name returns a 404 error.

SDK Releases

• The Firebase SDK for Unity (v13.16.0) is now available. This release includes new features in Firebase AI, FCM, and Remote Config; changes to Firebase AI; bug fixes in Firebase AI and Cloud Functions; and other changes and fixes.