GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531 —
CL-OP46 497 —
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531 —
CL-OP46 497 —
CL-OP48 490 -0.2%
← Back to feed

Google's Gemini API Managed Agents Switch to 3.6 Flash as Default — Add Tool Hooks, Budget Controls, Free Tier

Google DeepMind updated its Managed Agents product inside the Gemini API on July 28, switching the default model from the previous configuration to Gemini 3.6 Flash and adding a set of infrastructure controls that address the two main complaints from production agent developers: lack of observability into tool calls, and uncontrolled compute spend.

What Changed

Default model: Managed Agents now run on Gemini 3.6 Flash out of the box. Gemini 3.6 Flash launched earlier this month at $1.50 per million tokens with 1M context and integrated computer use. It posts 49% on DeepSWE and leads Terminal-Bench 2.1 at 76.2%, making it the fastest frontier-class coding model currently available from Google.

Environment hooks: The new feature lets developers intercept tool calls before they execute inside the sandbox. Hooks can block a call, lint it against a schema, or push it to an audit log. This closes a gap that mattered for compliance: previously, there was no way to inspect or gate individual tool invocations without rewriting the agent loop.

Budget controls: Per-session compute caps prevent runaway agent spending. Previously, a long-running agent could exhaust quota without a circuit breaker. The new controls let teams set a token or cost ceiling per session.

Scheduled triggers: Agents can now be triggered on a schedule rather than only on demand. Combined with the existing background task support, this enables durable periodic agents that don’t require a persistent client connection.

Free tier: Managed Agents in the Gemini API now has a free access path. The previous setup required a paid plan.

Context: What Managed Agents Actually Does

A single Gemini API managed agent call executes a complete agentic loop inside an isolated cloud sandbox: reasoning, code execution, package installation, file management, and web retrieval. Developers don’t wire these components themselves — the sandbox handles the orchestration. The prior release added background tasks and remote MCP server integration.

The hooks addition is the most operationally significant change. Enterprise AI teams deploying agents at scale have consistently cited tool call transparency as a prerequisite for production deployment. Anthropic added similar capabilities to Claude Managed Agents’ public beta earlier this month; Google’s hooks implementation gives its product parity on that dimension.

Positioning

Google is competing on infrastructure depth more than raw model capability at the agent layer. With Grok 4.5 in Copilot, Claude in AWS AgentCore, and OpenAI’s own agent runtime, the orchestration layer is becoming as contested as the model layer. The free tier addition suggests Google is prioritizing developer adoption over near-term monetization of the runtime itself.