Gemini 3.6 Flash default model, custom subagent instructions, and Pydantic argument coercion

Tuesday, 21 July 2026

AI summary

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

The Google Antigravity Python SDK has upgraded to Gemini 3.6 Flash and added several new features for improved functionality.

Default model upgraded to Gemini 3.6 Flash
Support for custom subagent instructions added
Automatic Pydantic argument coercion implemented
Defensive prompt sanitization to prevent errors
Configurable tool retries for better performance

Why it matters: Developers using the Google Antigravity SDK should care for enhanced model capabilities and improved error handling.

Official notes on antigravity.google

Full release notes14 changes

Release 0.1.8 of the Google Antigravity Python SDK updates the default text model to Gemini 3.6 Flash, adds support for custom subagent instructions, and implements automatic Pydantic argument coercion for tool calls. It also brings defensive prompt sanitization, configurable tool retries, and comprehensive stability improvements for local agent execution.

Default Model Upgrade to Gemini 3.6 Flash: Upgrades the default generative text model in the Python SDK to gemini-3.6-flash.
Pydantic TypeAdapter Tool Argument Coercion: Automatically coerces stringified numbers, booleans, and nested models returned by LLMs into strict Python types declared in tool signatures.
Custom Subagent Instructions: Adds capability to specify custom system instructions and allowlisted tool configurations for spawned subagents in multi-agent workflows.
Prompt Sanitization: Strips null bytes (\x00) and non-printable control characters (DEL, BEL, C1) from incoming user prompts at the wire boundary to prevent HTTP 400 errors and terminal corruption.
Configurable Tool Retries: Add RetryConfig definitions in local configuration to allow fine-tuned model and output retry policies.
Operating System Telemetry: Populate client OS and version information in telemetry headers to improve diagnostic tracking.
Native Usage Accumulation: Enable native addition (+ and +=) operations on UsageMetadata instances.
Large Tool Output Handling: Dynamic output truncation and removal of WebSocket frame limits to resolve connection termination on large tool responses.
Subagent Idle Synchronization: Fix race conditions where multiple idle states could cause receive_steps() to hang indefinitely.
Custom Tool Policy Enforcement: Ensure custom tools properly trigger pre-tool policy checks and cleanly report denials.
Media MIME Type Inference: Raise ValueError instead of Pydantic validation failures when media MIME types cannot be inferred from file extensions.
LiteRT Log Noise: Suppress verbose C++ LiteRT engine diagnostic output by default.
Policy Copy Idempotency: Prevent duplicate workspace policy prepending during deep copies of BaseLocalAgentConfig.
Type Safety: Enforce typed types.Step arguments in OnCompactionHook.

Full release notes14 changes

Default Model Upgrade to Gemini 3.6 Flash: Upgrades the default generative text model in the Python SDK to gemini-3.6-flash.
Pydantic TypeAdapter Tool Argument Coercion: Automatically coerces stringified numbers, booleans, and nested models returned by LLMs into strict Python types declared in tool signatures.
Custom Subagent Instructions: Adds capability to specify custom system instructions and allowlisted tool configurations for spawned subagents in multi-agent workflows.
Prompt Sanitization: Strips null bytes (\x00) and non-printable control characters (DEL, BEL, C1) from incoming user prompts at the wire boundary to prevent HTTP 400 errors and terminal corruption.
Configurable Tool Retries: Add RetryConfig definitions in local configuration to allow fine-tuned model and output retry policies.
Operating System Telemetry: Populate client OS and version information in telemetry headers to improve diagnostic tracking.
Native Usage Accumulation: Enable native addition (+ and +=) operations on UsageMetadata instances.
Large Tool Output Handling: Dynamic output truncation and removal of WebSocket frame limits to resolve connection termination on large tool responses.
Subagent Idle Synchronization: Fix race conditions where multiple idle states could cause receive_steps() to hang indefinitely.
Custom Tool Policy Enforcement: Ensure custom tools properly trigger pre-tool policy checks and cleanly report denials.
Media MIME Type Inference: Raise ValueError instead of Pydantic validation failures when media MIME types cannot be inferred from file extensions.
LiteRT Log Noise: Suppress verbose C++ LiteRT engine diagnostic output by default.
Policy Copy Idempotency: Prevent duplicate workspace policy prepending during deep copies of BaseLocalAgentConfig.
Type Safety: Enforce typed types.Step arguments in OnCompactionHook.