OpenAI’s official Python library for building applications that use the OpenAI API. Useful for developers who want to authenticate and make API requests from Python.
Project status
- The
openai/openai-pythonsource appears actively maintained, with the latest upstream push on 2026-10-15 and multiple SDKupdatesshipped in the last two weeks. - Update cadence looks fast and incremental, with new versions arriving roughly every few days (for example v3.23.0 on 10-01, v3.24.0 on 10-02, then several updates through 10-14).
AI summary generated
Recent updates
v3.35.1
v3.35.1 is a small patch release that focuses on a bug fix for beta Realtime websocket byte handling with websockets 13. The only runtime code change is in the beta realtime `recv_bytes()` implementation, with additional test coverage updates around websocket redirect behavior.
v3.35.0
v3.35.0 adds typed Live sideband server event models, including reflected input and output audio plus transport notifications. It also extends Agent service tier typing to include a new `ultrafast` tier, and fixes realtime WebSocket byte handling for websockets 13.
Featuresv3.34.0
v3.34.0 adds two outward-facing API capabilities: a new Realtime Translation WebSocket sessions feature, and a new degraded external storage status value. The code diff also introduces a new Translation session helper implementation plus additional WebSocket internals updates that are not covered by the release notes.
Featuresv3.33.0
v3.33.0 adds support for safety identifiers on image generation and editing requests, and surfaces the effective safety identifier on image-related response models and events. It also introduces a shared typed definition for the Decisions prompt_cache_breakpoint parameter and updates the input parameter types to reference it.
Featuresv3.32.0
v3.32.0 adds new API capabilities including prompt-based voice creation, prompt caching options, and new Live session features for audio fixture playback and Decisions. It also improves realtime/error diagnostics and refactors websocket connection handling to share logic across clients. The diff shows multiple API surface and typing changes beyond what the release notes explicitly spell out.
Featuresv3.31.0
v3.31.0 introduces the first set of new browser and computer toolset support in the SDK (labeled as part 1 of 4). The diff primarily adds new documentation and example/demo code for browser automation and integrates Playwright-based type checking and security checks into CI.
Featuresv3.30.0
v3.30.0 adds beta support for managed session spending limits, OCI external storage provider types, and organization Decisions usage reporting types. It also extends image-related beta response models with structured moderation failure details, and updates documentation around Decisions image inputs to allow public HTTP(S) image URLs.
Featuresv3.29.0
v3.29.0 expands the Agents beta surface area (vault updates, new metadata filters, and additional pagination parameter constraints) and adds two new end-user visible API elements, `billing_not_active` session error code and the built-in `shida` live voice. It also updates multiple documentation strings around Decisions and increases the documented maximum length for safety identifiers to 128 characters.
BreakingFeaturesv3.28.0
v3.28.0 adds support for agent session environment suspension and expiration. It introduces new beta event models and extends environment status enums and webhook event typing so clients can parse and deliver the new lifecycle events.
SecurityFeaturesv3.27.0
v3.27.0 adds beta support for prewarmed OpenAI-hosted agent environments. The change introduces new API methods to create and list environments, and adds corresponding webhook event types for environment lifecycle notifications.
Featuresv3.26.1
v3.26.1 is a small release focused on audio “custom voice” creation. However, the code diff shows a broader schema and API surface change that removes support for prompt-based voice creation.
v3.26.0
v3.26.0 primarily adds standalone Decisions support to the client, including a new top-level `decisions` resource and a set of generated type definitions for Decision inputs and responses. In addition to the Decisions work, the release also introduces new tool-search related types (including a new tool namespace wrapper and more MCP-related tool schema), but these changes are not mentioned in the release notes.
Featuresv3.25.0
v3.25.0 adds new Agents streaming behavior for local tool execution and improved diagnostics for tool failures. It also extends admin organization usage reporting with grouping by usage source (api_source) and exposes additional library helper functionality.
Featuresv3.24.0
v3.24.0 adds API support for creating custom audio voices and exposes new audio.voices resources. It also includes bug fixes related to prioritizing routing fields during Python request routing and parsing.
Featuresv3.23.0
v3.23.0 adds several beta Agents capabilities, including typed answer collection from session event streams, staging files for hosted agent environments, and downloading artifacts from specific completed turns. It also introduces API/session improvements such as session traces and Realtime translations, plus multiple bug fixes around reconnect behavior and endpoint/path query preservation.
Featuresv3.22.1
v3.22.1 includes a fix to the client authentication error message and updates the response transformation logic to correctly handle TypedDict fields wrapped with NotRequired. The code diff also shows additional parsing robustness improvements not mentioned in the release notes.
v3.22.0
v3.22.0 adds support for “computer use” in beta agent flows, including new browser authentication and computer use call output types. The codebase also introduces new hosted-environment configuration for provisioning a desktop and a new network restriction field.
Featuresv3.21.0
Release v3.21.0 primarily adds support for a new model identifier, "gpt-6.1-sol", across the SDK. The change is reflected in the various typed parameter definitions and the Responses/Beta Responses resource methods.
Featuresv3.20.0
v3.20.0 adds new Responses/Agents capabilities including Cyber access program controls and a new opt-in WebSocket accumulator for incremental text and tool-input snapshotting. It also contains Live and client robustness fixes around WebSocket behavior, TLS retry behavior, and fractional transcript grouping deadlines.
Featuresv3.19.2
v3.19.2 includes a targeted bug fix in file extraction fallback behavior, plus documentation/spec clarifications around web search user location defaults and some API schema details. The code diff is mostly schema/docstring updates, with one meaningful change in internal file extraction logic.