Google DeepMind says Managed Agents in the Gemini API now use Gemini 3.6 Flash by default, while adding environment hooks, budget controls, scheduled triggers, model selection and free-tier access.
Google DeepMind has updated its Managed Agents offering in the Gemini API, making Gemini 3.6 Flash the default model for its Antigravity managed agent and adding controls intended for supervising work performed in an isolated sandbox.
In a July 28, 2026 post, Google said the release introduces environment hooks, model selection, budget controls, scheduled triggers and free-tier availability. The changes extend earlier Managed Agents capabilities, including background tasks and integrations with remote Model Context Protocol (MCP) servers.
According to Google’s announcement, the antigravity-preview-05-2026 agent now runs on Gemini 3.6 Flash by default. Google says developers do not need to make code changes to receive that default-model update.
Google’s Agents Overview documentation identifies Antigravity as a managed agent powered by Gemini 3.6 Flash, while also documenting configurable model options, sandboxing, pricing and free-tier access.
Managed Agents are available through the Gemini Interactions API. Google describes an interaction as a single API call that can coordinate reasoning, code execution, package installation, file management and web retrieval within an isolated cloud sandbox. That structure is designed to let developers delegate multi-step work while using Google-managed infrastructure for the execution environment.
The new environment hooks are intended to give developers a way to inspect or control tool calls made inside the sandbox. Google says hooks can be used to block, lint or audit those calls.
The feature could be relevant for teams that need policy checks around actions such as installing packages, accessing files or retrieving information from the web. Google’s announcement does not position hooks as a replacement for application security controls; rather, they are an additional mechanism within the managed execution environment.
Google also announced budget controls for Managed Agents. Its Antigravity Agent documentation describes a max_total_tokens setting that can be used to constrain the total token budget for an Antigravity interaction.
The company said scheduled triggers are part of the update, allowing Managed Agents to be initiated on a schedule. Google also said the service is available on both free and paid tiers, a detail corroborated by its official product documentation.
For developers using JavaScript or TypeScript, Google’s announcement provides examples based on the @google/genai SDK. It directs Python and cURL users to the Antigravity agent documentation.
The release combines a default-model change with more operational controls around autonomous, tool-using workloads. Gemini 3.6 Flash supplies the default reasoning model, while hooks and token budgets offer mechanisms for governing execution and spend. Scheduled triggers broaden the ways these workflows can be started, and free-tier availability lowers the barrier to experimenting with the service.
Google’s announcement and documentation characterize Antigravity as a preview offering, so developers should review the current product documentation for availability, model configuration and pricing details before deploying it in production.
In a July 28, 2026 post, Google said the release introduces environment hooks, model selection, budget controls, scheduled triggers and free tier availability.
The changes extend earlier Managed Agents capabilities, including background tasks and integrations with remote Model Context Protocol (MCP) servers.
Gemini 3.6 Flash becomes the default According to Google’s announcement, the antigravity preview 05 2026 agent now runs on Gemini 3.6 Flash by default.
Continue reading
