npx @buildinternet/releases get openai-python-sdkIn Live response models, question instructions can now be a string or a list of structured text, image, or JSON parts, and Decisions request input types include additional message roles and tool-history items. Existing string instructions and DecisionInputMessageParam user-message inputs remain supported.
Fixed the API client not honoring per-call WebSocket query settings, and browser calls are now validated before reading driver state. Also stopped microphone capture between recordings in examples.
The Responses API now supports Vertex AI, alongside new 30-minute cache-write usage counts, compaction trigger IDs, and custom voice consent management. Fixed issues in WebSocket setup and close handling, Vertex API root preservation, and Pydantic v1 alias validation.
Fixed beta Realtime byte reads on websockets 13, which were previously returning unexpected data on that client version.
Added typed Live sideband events and an ultrafast service tier for Agents. Fixed realtime byte receives being dropped with websockets 13.
Adds Realtime Translation WebSocket sessions and a degraded external storage status to the API surface.
Image requests and responses now include safety identifiers, and the Decisions cache breakpoint parameter type is shared across the API.
The API now supports prompt-based voice creation, cache settings and key options, regional processing residency options, and response error diagnostics with model notices. Live sessions gain audio fixture playback and Decisions support, and Realtime response error messages are now declared.
Adds browser and computer toolsets as the first of a four-part series.
Adds beta session spending limits and organization Decisions usage reporting, plus OCI external-storage provider support, image URLs with detailed safety alert explanations, and the ability to update project API key expiry. Fixed async HTTP/1.1 connection reuse in streaming.
Vault updates and metadata filters are added to the API, along with a billing_not_active session error code and the shida named Live voice. Beta pagination limits and non-null parameters are aligned.
The API now supports suspending and expiring agent environments. PyJWT security constraint updated to 2.14.0.
The API now supports prewarmed hosted environments.
Fixed incorrect parameters in the custom voice creation API call.
The API now supports standalone Decisions.
Agents can now observe local tool failures and run local tools during streamed session creation, and typed function helpers support tool search with defer_loading=true. Also added agent turn items and usage source grouping, and exported type_to_response_format_param publicly.
Adds custom voice creation and agent session events to the API. Also fixes Python request routing fields to be prioritized in requests and parse helpers.
The API now exposes session traces and Realtime translations, and Agents can stage files and download turn artifacts. Fixed a Chat Completions issue with original image detail, corrected the eval run cancellation endpoint, retained WebSocket endpoint paths and query parameters, and hardened Responses reconnect behavior against replaying uncertain typed sends.
Fixed handling of NotRequired typed dictionary fields and corrected the missing authentication error message. Documentation and test chore updates accompany the release.
The SDK adds computer use support to beta agents.
