Google Defaults Gemini API Managed Agents to 3.6 Flash, Adds Environment Hooks and Free Tier
Gemini's managed agents now run on Gemini 3.6 Flash by default, with new environment hooks for tool-call auditing, budget controls, and free tier access.
Last verified:
Gemini 3.6 Flash Becomes Default for Managed Agents
According to Google AI Blog, the Antigravity agent now runs on Gemini 3.6 Flash by default, with no code changes required for existing integrations. Developers can also explicitly select alternative models—Gemini 3.5 Flash, Gemini 3.5 Flash-Lite for cost reduction, or a pinned model of their choice—by passing the agent_config.model parameter when creating an interaction. This flexibility addresses cost-conscious and latency-sensitive use cases without forcing a single-model constraint.
Environment Hooks Enable In-Sandbox Compliance
Google AI Blog introduces environment hooks, a new mechanism for running custom scripts before or after every tool call within the managed agent’s sandbox. Developers add a .agents/hooks.json configuration file to specify handlers for pre_tool_execution and post_tool_execution events, with regex matchers to target specific tools. This design allows teams to block risky function calls, lint code before execution, and audit all tool invocations—critical capabilities for regulated industries and security-conscious deployments that previously relied on API-level middleware.
Budget Controls and Scheduled Execution
According to Google AI Blog, the update adds budget controls to cap execution costs and scheduled triggers to enable background job execution. These features address two operational pain points: cost runaway in long-running agentic workflows and the need to offload tasks that do not require synchronous user interaction. Combined, they reduce operational friction for production deployments.
Free Tier Removes Cost Barrier to Adoption
Google AI Blog confirms that free tier access is now available for managed agents, lowering the initial investment required to prototype and evaluate the platform. This signals Google’s intention to compete on experimentation velocity rather than gate-keeping behind paid tiers, matching similar moves by competitors offering free-tier agent runtimes.
Why This Matters
Teams evaluating agentic platforms must now weigh three distinct architectures: Gemini’s managed sandbox (with in-process environment hooks for compliance), OpenAI’s Agents (stateless API-only execution), and Anthropic’s MCP protocol (agent-agnostic tool abstraction). The introduction of environment hooks and free tier access shifts the decision axis from pure cost to compliance maturity and operational control. Organizations subject to audit or data-residency requirements now have a sandboxed execution environment with auditable tool calls—a material advantage over cloud-only competitors. Developers prioritizing rapid prototyping will benefit from the free tier; cost-conscious teams will choose Gemini 3.5 Flash-Lite; teams requiring governance will layer environment hooks.
Frequently Asked Questions
What is Gemini's managed agents feature?
Managed agents in the Gemini Interactions API coordinate reasoning, code execution, package installation, file management, and web retrieval inside an isolated cloud sandbox through a single API call.
Can I use a different model instead of Gemini 3.6 Flash?
Yes. According to Google AI Blog, developers can explicitly select Gemini 3.5 Flash, Gemini 3.5 Flash-Lite, or pin to a model of preference by passing agent_config.model when creating an interaction.
What are environment hooks and why do they matter?
Environment hooks let developers run custom scripts before or after each tool call the agent makes. They support regex matching and enable blocking, linting, and auditing of tool calls inside the sandbox—critical for compliance and security workflows.
Is there a free tier for managed agents?
Yes. Google AI Blog confirms that free tier access is now available for managed agents, lowering the barrier to entry for experimentation and small-scale deployments.