Default Gemini 3.8 Flash model upgrade, lightweight agent configuration, and Vertex AI Express mode

Monday, 31 August 2026

AI summary

Written by AI from the official notes. Check them for exact details.

The default model has been upgraded to Gemini 3.8 Flash, and new features for lightweight configurations and Vertex AI Express mode have been added.

Default model updated to gemini-3.8-flash for better reasoning and task performance.
Introduced .lightweight() method for optimized small and local agent configurations.
Added Vertex AI Express mode for simplified API key authentication.
Enhanced ToolRunner support for callable classes and dataclass methods.
New enable_sandbox field for executing commands in an OS-level sandbox.

Why it matters: Developers using Google AI tools should care for improved performance and easier configurations.

Official notes on antigravity.google

Full release notes12 changes

The 0.1.16 release updates the default model for new agents to gemini-3.8-flash for enhanced reasoning capabilities, introduces a .lightweight() method on agent configs for small and local models, adds Vertex AI Express mode with API keys, and expands tool runner support for callable classes and dataclass functors.

Default Model Update to Gemini 3.8 Flash: The default model for new agents and configurations has been updated to gemini-3.8-flash, delivering higher reasoning quality and stronger task performance. Developers can override this default by specifying model in their configuration:
Optimized Lightweight Agent Configuration: New .lightweight() method on agent configurations (including LocalAgentConfig and LiteRTAgentConfig) applies preset optimizations for smaller, local-running models by configuring minimal tool sets, minimal prompting, and disabling background subagents:
Vertex AI Express Mode (API Key Support): Developers can now connect to Vertex AI using Express mode by providing an API key directly on LocalAgentConfig(vertex=True, api_key="..."), simplifying authentication without needing full GCP project/location ADC setup:
Support for Callable Class and Dataclass Tools: The SDK's ToolRunner now inspects and executes tools defined as callable class instances (functors) and dataclass methods, enhancing flexibility for custom tool implementations:
OS Sandbox Opt-in for Commands: Added an enable_sandbox field to RunCommandConfig to allow developers to execute terminal commands within an OS-level sandbox environment:
Tool Invocation Argument Flexibility: ToolWithSchema and public callable proxies now accept both positional ( args) and keyword arguments ( *kwargs), matching standard Python function calling conventions.
Support for genai.Content Media: Introduced a structconverter to support media blocks from genai.Content objects within the SDK, enabling richer multimodal interactions.
MCP Dependency Compatibility Widening: The SDK now supports both mcp>=1.0 and mcp<3.0 dependencies.
Stop Hook Integration: Agents now support the StopHook lifecycle hook, triggered when an agent's execution is externally stopped, enabling custom cleanup and resource logging actions:
Interactive Example Agent Behavior: Fixed interactive SDK examples (interactive_cli.py, human_in_the_loop.py) to consistently configure AgentBehavior.INTERACTIVE.
Tool Runner stdio MCP Environment: Resolved an issue in exported SDK examples where stdio MCP servers failed by ensuring child processes use sys.executable within the active virtual environment.
Step Token Usage Deprecation: Deprecated Step.usage_metadata in favor of consolidated turn-level (ChatResponse.usage_metadata) and session-level (agent.conversation.total_usage) reporting.

Full release notes12 changes

Default Model Update to Gemini 3.8 Flash: The default model for new agents and configurations has been updated to gemini-3.8-flash, delivering higher reasoning quality and stronger task performance. Developers can override this default by specifying model in their configuration:
Optimized Lightweight Agent Configuration: New .lightweight() method on agent configurations (including LocalAgentConfig and LiteRTAgentConfig) applies preset optimizations for smaller, local-running models by configuring minimal tool sets, minimal prompting, and disabling background subagents:
Vertex AI Express Mode (API Key Support): Developers can now connect to Vertex AI using Express mode by providing an API key directly on LocalAgentConfig(vertex=True, api_key="..."), simplifying authentication without needing full GCP project/location ADC setup:
Support for Callable Class and Dataclass Tools: The SDK's ToolRunner now inspects and executes tools defined as callable class instances (functors) and dataclass methods, enhancing flexibility for custom tool implementations:
OS Sandbox Opt-in for Commands: Added an enable_sandbox field to RunCommandConfig to allow developers to execute terminal commands within an OS-level sandbox environment:
Tool Invocation Argument Flexibility: ToolWithSchema and public callable proxies now accept both positional ( args) and keyword arguments ( *kwargs), matching standard Python function calling conventions.
Support for genai.Content Media: Introduced a structconverter to support media blocks from genai.Content objects within the SDK, enabling richer multimodal interactions.
MCP Dependency Compatibility Widening: The SDK now supports both mcp>=1.0 and mcp<3.0 dependencies.
Stop Hook Integration: Agents now support the StopHook lifecycle hook, triggered when an agent's execution is externally stopped, enabling custom cleanup and resource logging actions:
Interactive Example Agent Behavior: Fixed interactive SDK examples (interactive_cli.py, human_in_the_loop.py) to consistently configure AgentBehavior.INTERACTIVE.
Tool Runner stdio MCP Environment: Resolved an issue in exported SDK examples where stdio MCP servers failed by ensuring child processes use sys.executable within the active virtual environment.
Step Token Usage Deprecation: Deprecated Step.usage_metadata in favor of consolidated turn-level (ChatResponse.usage_metadata) and session-level (agent.conversation.total_usage) reporting.