OpenAI Python SDK v3.1.0 Adds Ultrafast Tier Support, Deprecates Sora Video APIs

OpenAI shipped version 3.1.0 of its official Python client library on August 14, 2026, introducing support for a new "Ultrafast" processing tier and beginning the formal deprecation of the Sora video generation API.

The release, published to the openai-python GitHub repository, bundles four feature additions and one infrastructure cleanup, according to the project's changelog.

What changed

WebSocket stream IDs (#3612). The SDK now supports stream identifiers for WebSocket connections, allowing developers to track and correlate individual streams within a single connection more precisely.

Workload identity access token event (#3601). A new event type surfaces when a workload identity access token is issued, giving developers programmatic visibility into token issuance for authentication flows that rely on workload identity rather than static API keys.

Sora video API deprecation (#3610). OpenAI has marked its Sora video APIs as deprecated within the SDK. The changelog does not specify a sunset date or migration path, and OpenAI has not published pricing or timeline details for the deprecation.

Ultrafast tier, structured MCP/WebSocket errors (#3617). The largest change in this release bundles three additions: client support for a new "Ultrafast" processing tier, structured error objects for Model Context Protocol (MCP) interactions, and separated event handling for WebSocket-specific errors. OpenAI has not disclosed pricing for the Ultrafast tier or detailed which models or endpoints it applies to.

Stainless infrastructure removal. As a chore item, OpenAI removed attribution and build infrastructure tied to Stainless, the third-party SDK-generation platform historically used to scaffold OpenAI's client libraries. This suggests OpenAI is moving SDK maintenance in-house or onto a different toolchain, though the company has not commented publicly on the change.

What this means

This is a software development kit update, not a new model release — no model weights, benchmark scores, or context window changes are involved. Three items merit attention from developers building on OpenAI's API.

First, the Sora deprecation signals that any production system using OpenAI's video generation endpoints should begin migration planning now, even without a firm end-of-life date. Second, the "Ultrafast" tier is the first public signal of a new latency-optimized service class, though without pricing or model-scope details, its practical value is unclear until OpenAI documents it separately. Third, the removal of Stainless infrastructure is an internal tooling change unlikely to affect API behavior, but it may precede broader changes to how OpenAI generates and versions its client libraries going forward.

Developers should treat the Ultrafast tier and Sora deprecation as items to monitor in OpenAI's official API documentation rather than acting on the SDK changelog alone, since GitHub release notes rarely carry full technical specifications.