NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #203 most downloaded on PyPI
GenAI Python SDK
Latest stable
3.6.1
Last release 4 days ago
18 Nov 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 60 of the last 60 stable releases
12 versions withdrawn
withdrawn after publishing
2 years old
142 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
3.6.1 (2026-11-18) Bug Fixes Update sdk enum
environments: Introduce self-hosted BYOS mode, system_instruction, custom_config, and environment_key in SDKs and CLI
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Deprecate non-interactions names in google.genai.interactions.
$defs references per path in process_schema for recursive union schemas (8a9c9fd)Update README to reflect breaking changes in major version 3
XHIGH and xhigh to ThinkingLevel across GenAI SDKs and CLI. (300f12f)Add the Decisions API ( client.decisions.create ) to the GenAI SDKs
Add gemini-omni-1.2-flash and make it the gemini-api video default
gemini-omni-1.2-flash and make it the gemini-api video default (8e9cce3)proactive_video to ProactivityConfig for the Live API. (97d0678)xhigh thinking level to Interactions ThinkingLevel (0f29f4d)Default Gemini Developer API version is now v1 (was v1beta )
Default Gemini Developer API version is now v1 (was v1beta) (2d0f947)
v1 instead of v1beta.vertexai=True / enterprise=True are unaffected by this default change and continue to default to v1beta1.v1, notably: the gemini-flash-latest, gemini-flash-lite-latest and gemini-pro-latest aliases, and some -preview models. They exist only on v1beta. To keep using them, use a specific stable model version, or switch back to v1beta as shown below.caches) are not available on v1. The Gemini Developer API serves them only on v1beta. To keep using them, switch back to v1beta as shown below.v1, notably: gemini-flash-latest, gemini-flash-lite-latest and gemini-pro-latest. They exist only on v1beta. To keep using them, use a specific model version, or switch back to v1beta as shown below.GOOGLE_GENAI_API_VERSION (works for both Gemini Developer API and Enterprise / Vertex AI clients when no version is set in code):
# Restores the beta default ('v1beta' on Gemini Developer API, 'v1beta1' on Enterprise / Vertex AI)
export GOOGLE_GENAI_API_VERSION=beta
# Or set an explicit API version string on either platform (e.g. 'v1', 'v1beta', 'v1beta1')
export GOOGLE_GENAI_API_VERSION=v1betaapi_version in HttpOptions (takes precedence over GOOGLE_GENAI_API_VERSION):
from google import genai
from google.genai import types
client = genai.Client(
http_options=types.HttpOptions(api_version='v1beta')
)Default HTTP client is now httpx2 (was httpx), and httpx is no longer installed by default (51285c8)
httpx is no longer installed by default. httpx2 is now the required dependency and default HTTP client.httpx default via environment variable: Because httpx2 is installed by default, installing google-genai[httpx] alone only makes import httpx available; to switch the SDK's default client back to httpx without code changes, both install the httpx extra and set GOOGLE_GENAI_HTTP_CLIENT=httpx:
pip install "google-genai[httpx]"
export GOOGLE_GENAI_HTTP_CLIENT=httpxhttpx client in code: After installing google-genai[httpx], you can pass httpx.Client / httpx.AsyncClient directly via HttpOptions:
import httpx
from google import genai
from google.genai import types
client = genai.Client(
http_options=types.HttpOptions(
httpx_client=httpx.Client(),
httpx_async_client=httpx.AsyncClient(),
)
)httpx code to httpx2: In most cases, updating your imports with import httpx2 as httpx is sufficient.httpx2 uses the OS certificate trust store (via truststore / httpx2.create_ssl_context()) instead of certifi when SSL_CERT_FILE and SSL_CERT_DIR are not set.httpx vs. httpx2 types matter: Package-specific objects passed in HttpOptions(client_args=..., async_client_args=...) (such as Timeout, Limits, MockTransport / BaseTransport, or custom Auth subclasses) and unwrapped transport exceptions caught from client.models, client.files, etc. (TimeoutException, ConnectError, HTTPError) must match the active HTTP client. (When GOOGLE_GENAI_HTTP_CLIENT is unset or auto, passing httpx objects in HttpOptions will also auto-select the httpx backend if google-genai[httpx] is installed.)interactions: CodeExecutionResultStep.result changes from string to the CodeExecutionResultStepResult union. Code that reads result as a string must now check which variant is present and read the string from it. The JSON for text-only results is unchanged.
source_text to AudioTranscriptionConfig and SmartTranscriptionMode. (1ac3c91)Add support for continuation_token and requires_continuation status
Add store to GenerateContentConfig (Gemini Developer API only) and labels to LiveConnectConfig .
store to GenerateContentConfig (Gemini Developer API only) and labels to LiveConnectConfig. (16a29c9)Add base_voice to PromptedVoice in GAOS SDK
send_context_update) and types across GenAI SDKs. (364889a)fixes.openSingleValueEnumsOct2026 across all SDKs (78f6d34)Any-typed fields (4c1abcc)interactions: Add gemini-4-argon to Model enum
Concatenate output_text across multi-step model responses in a turn
Accumulate streaming continuation usage metadata and expose initial_prompt_token_count
continuation_token to the Interactions interaction.status_update event. (02db4c5)super().model_construct in Interaction model validator (ccd5dd3)Support "allowlist": "disabled" in environment network config.
Add lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to Interactions Model enum
Support continuation_token in GenerateContent
Add gcs_metrics_uri for exporting tuning job metrics.
--usage schema (2579b94)CreateAgentInteraction for Trigger.interaction and TriggerCreateParams.interaction (046ccfb)Add sample_audio to Voice in GAOS SDK
Add credential APIs resource to gaos.
Add from_environment support for environment copying
Add Gemini 3.8 Flash model to SDKs and update Flash model descriptions
Add destination parameter to client.files.download to support streaming downloads to disk or file-like objects
Add AudioTranscriptionConfigMode
Add IDLE state to live connection status enum and mark REQUIRES_ACTION as deprecated.
mode enum (VERBATIM, SMART) to AudioTranscriptionConfig and TranscriptionConfig. (e8cd7d0)## 2.18.1 (2026-08-13) ### Bug Fixes * Add gemini-3.7-flash
Add interaction_status to LiveServerContent
Add the Gemini Robotics ER 2 Preview model
Support GoogleMaps Tool grounding_types places and routing
[GenerateContentConfig] Add GenerationConfig.audio_transcription_config and Part.audio_transcription.
language_codes field to AudioTranscriptionConfig. (fc282b3)Add deprecation warnings to Imagen generate_images, edit_images, generate_videos (if using prompt/text/image args), LiveConnectConfig.GenerationConfig…
A new field custom_vocabulary is added to to Live API
custom_vocabulary is added to to Live API (4eeb1ad)custom_vocabulary is added to message .google.cloud.aiplatform.v1beta1.BidiGenerateContentSetup (4eeb1ad)Fix the normalization of Interactions inside triggers
Add AntigravityAgentConfig to OpenAPI schema overlays and regenerate SDKs.
Add environment_id in network config
Add Agent Platform MCP support to async generate_content_stream
Use .model_copy() instead of deprecated .copy()
aiohttp.readline without max_line_length for backward compatibility because we still want to keep aiohttp as optional dependency (e99ab99), closes #2487transformers an optional dependency for local tokenizers, also add other dependencies to local-tokenizer-extras. (528926b)Add audioOffset to VoiceActivity
Announce Automatic Function Calling (AFC) upcoming breaking change warning
aiohttp.readline without max_line_length for backward compatibility because we still want to keep aiohttp as optional dependency (e99ab99), closes #2487transformers an optional dependency for local tokenizers, also add other dependencies to local-tokenizer-extras. (528926b)Add Agent Platform MCP support to async generate_content
enable_widget in message GoogleMaps is changed (74d81dd)google_maps_widget_context_token in message GroundingMetadata is changed (74d81dd)codegen_instructions.md for simpler maintenance, Gemini API Skills should be the single source of truth (bfa2a49)Additional computer_use field support for vertex.
Add enable_prompt_injection_detection for Computer Use feature for the Gemini API.
Add Gemini 3.5 Flash model to options
Support Agent and Environment APIs.
Add content union to UserInputStep
Added missing FunctionCallResultDelta type and arguments field to the ArgumentDelta type
Add gemini-3.1-flash-lite to model options
Update response_format field names to snake_case.
Note: The breaking changes are only in interactions. GenerateContent usage in unaffected. Refer to https://ai.google.dev/gemini-api/docs/interactions-…
Note: The breaking changes are only in interactions. GenerateContent usage in unaffected.
Refer to https://ai.google.dev/gemini-api/docs/interactions-breaking-changes-may-2026 for details
[Python] Multimodal file search
[Interactions] Add FileCitation.{custom_metadata,media_id,page_number}
output_info to BatchJob (7b77ab8)enterprise to Client constructor and GOOGLE_GENAI_USE_ENTERPRISE (693fd9a)AsyncClient.__aexit__, __exit__. (a74dc65)Vertex AI with Gemini Enterprise Agent Platform, update method error message to replace Vertex AI with Gemini Enterprise Agent Platform (previously known as Vertex AI), update converter error message to replace Vertex AI with Gemini Enterprise Agent Platform (413f0f9)Refactor Webhook types in GenAI SDKs for easier useage
> CRITICAL WARNING: Do not use this version if you are implementing or relying on webhooks. This release contains known issues regarding webhook sdk.
[!CAUTION] CRITICAL WARNING: Do not use this version if you are implementing or relying on webhooks. This release contains known issues regarding webhook sdk. Please use v1.74.0 or later.
Add "eu" as a supported service location for Vertex AI platform.
Remove deprecated product recontext model samples from docstrings
Support dedicated TextAnnotationDelta for streaming tool responses
Add consent_audio and voice_consent_signature and AsyncSession.setup_complete
[Interactions] Breaking change to Interactions API to refactor TextContent annotations to use specific citation types
attempts=0 as attempts=1 in retry options to ensure no retries (2856c0a)> This release contains a bug where the typing-extensions lower bound is set too low, which causes the SDK to break in some environments.
[!CAUTION] Known Issue: This release contains a bug where the
typing-extensionslower bound is set too low, which causes the SDK to break in some environments.Recommended Actions:
- Fall back to 1.66.0: If you do not immediately need the new features in 1.67.0, we recommend sticking with the previous stable version:
pip install google-genai==1.66.0- Manual Fix: If you must use 1.67.0, you can resolve the issue by manually upgrading
typing-extensions:pip install "typing-extensions>=4.14.0"
Add gemini-3.1-flash-image-preview model
Change interactions media mime type to enum (breaking change for experimental feature)
One column per month.