Google on July 28 expanded its Gemini API Managed Agents platform with a set of controls aimed squarely at production deployments: a smarter default model, programmable environment hooks, token budget caps, scheduled execution triggers, and access to the free tier, all without requiring developers to change existing code.
Gemini 3.6 Flash becomes the new default
Google has updated its Gemini API Managed Agents with Gemini 3.6 Flash as the default model, and the Antigravity preview agent now uses it automatically, with no code changes required from developers already using the service.
Developers can also select another supported model through the agent configuration. Available options include Gemini 3.5 Flash and Gemini 3.5 Flash-Lite. Google positions Gemini 3.5 Flash-Lite as the fastest, lowest-cost model in the 3.5 family, optimized for high-throughput execution. Developers can pass agent_config.model when creating an interaction or managed agent to specify a preferred model explicitly.
Gemini 3.6 Flash is Google’s workhorse Gemini 3 model for agentic workflows, coding, knowledge work, and multimodal tasks. Built on Gemini 3.5 Flash, it improves task quality while using fewer reasoning steps, tool calls, and output tokens.
Output tokens cost $7.50 per million on 3.6 Flash, down from $9.00 per million on 3.5 Flash, while input tokens remain at $1.50 per million. On the DeepSWE benchmark, Gemini 3.6 Flash scores 49% versus 37% for Gemini 3.5 Flash.
Environment hooks give developers fine-grained control
Environment hooks allow custom scripts to run before or after tool calls within the sandbox. That means developers can intercept an agent’s actions at the tool-call boundary to enforce validation rules, trigger external pipelines, or log outputs to a remote endpoint, all from inside the isolated execution environment.
The features build on a July 7, 2026 release that gave managed agents long-running background task support and remote Model Context Protocol (MCP) server integration.
Requests sent through a post-execution hook travel through Google’s egress proxy, so the destination must sit on the environment’s allowlist, and authentication tokens live in the network configuration rather than in the hook file. This design keeps credentials out of agent code, a meaningful guard for teams running agents against production systems.
For developers building AI assistant pipelines at scale, the hook architecture reduces the need for external orchestration layers.
Token caps and scheduled triggers round out the release
Because managed agents execute multi-turn autonomous loops, complex tasks can consume significant token budgets. To prevent runaway tasks, developers can pass max_total_tokens inside agent_config to cap total consumption across input, output, and thinking.
When the agent reaches the limit, execution safely pauses and the interaction returns a status of “incomplete.” The environment state is preserved, enabling the developer to continue where it stopped by passing previous_interaction_id with a fresh budget.
Scheduled triggers bind an agent, a prompt, an environment, and a cron expression into a persistent resource that fires on its own, pausing itself by default after five consecutive failures. Every run reuses the same sandbox, so files written by one execution are visible to the next.
An environments interface also lets developers list, inspect, and delete sandbox sessions from code, which recovers an environment ID after a dropped connection and clears sandboxes when a pipeline finishes instead of waiting out their seven-day expiry.
Free tier opens the platform to more developers
Managed agents are now available on free tier projects. Developers can experiment with agentic workflows using an API key from a project without active billing.
The Interactions API reached general availability on June 22, 2026, and is the primary interface for Gemini-based agent development. The combined effect of the free tier and the new budget controls lowers the barrier for smaller teams to move agents from prototype to production.
Google’s update positions Managed Agents as infrastructure suitable for recurring, automated, and high-stakes workloads, not just exploratory one-off tasks.
