Deprecations
Open official pageOpenAI deprecations
Source of record for the Realtime API beta deprecation notice, shutdown date, and replacement path.
OpenAI deprecation detail
Realtime API beta shutdown date: May 7, 2026.
This page answers the Realtime API beta shutdown query directly. It keeps the shutdown date, GA replacement path, protocol changes, and cutover checks in one source-linked view.
Why this matters now
The beta Realtime path now has a hard stop.
Use these points to decide whether the team should stop new beta work, schedule the GA migration, or re-test the current voice stack before it becomes a deadline problem.
Official sources
Check the OpenAI pages behind this migration brief.
This page uses a fixed source set so the migration brief stays auditable instead of drifting into secondhand advice.
Guide
Open official pageRealtime guide
Includes the Beta to GA migration section and the current GA interface guidance for Realtime sessions, events, and item handling.
Interface mapping
The GA path changes the transport contract your client has to handle.
The migration work is about the wire format and session behavior as much as the endpoint. These are the shifts the OpenAI guide flags before beta traffic can move cleanly to GA.
| Before | Now | What changes | Sources |
|---|---|---|---|
| Beta connection setup | GA Realtime API session setup | The beta interface is treated as a separate integration path in OpenAI's migration guidance. Teams should rebuild connection setup against the GA Realtime API instead of keeping a beta-specific bootstrap path alive. | |
| Beta event names and shapes | GA event contract | OpenAI calls out new event names and shapes in GA, which means client-side dispatchers, validators, and test fixtures need explicit updates rather than a pass-through swap. | |
| Beta input and output item handling | GA input and output item model | The guide flags item-level changes in GA, so streaming logic, transcript handling, and any stored event normalization should be validated against the new item model. | |
| Beta client assumptions | GA cutover checks | OpenAI publishes a dedicated Beta to GA migration section because a production move requires interface validation across sessions, responses, and realtime event flow before traffic is switched. |
Migration checklist
Use the shutdown date to force the GA move into sequence.
The order here is meant to stop fresh beta debt first, then move the protocol surface and production test coverage under control.
Step 1
Stop net-new beta integrations.
Treat the beta interface as a migration source only. New realtime work should be scoped against the GA Realtime API rather than extending beta-specific setup, events, or handlers.
Step 2
Audit beta-specific connection and event handling.
Find where the client still assumes beta setup, beta event names, or beta event shapes. Those assumptions need to be replaced deliberately because OpenAI calls out contract changes in GA.
Sources
Step 3
Rebuild staging flows on the GA item model.
Update staging clients to the GA Realtime API and verify that session lifecycle, streaming outputs, and any input or output item handling match the current guide.
Sources
Step 4
Re-test voice and interruption flows before cutover.
Realtime migrations are only finished when the team has replayed common voice turns, interruptions, retries, and failure handling on the GA path instead of assuming transport compatibility.
Sources
Verify before cutover
Do not treat the GA switch as a header or endpoint edit.
These checks are the minimum confidence gates before production voice or realtime traffic should leave the beta interface.
Connection and session setup
Validate authentication, session initialization, and transport setup against the current GA guide instead of carrying beta bootstrap code forward by assumption.
Sources
Event parser and dispatcher coverage
Confirm the client can parse and route the GA event names and event shapes OpenAI documents, especially around session updates, response creation, and streamed output items.
Sources
Streaming, transcript, and item handling
Re-run the paths that persist or interpret input and output items so transcript assembly, UI updates, and downstream analytics still match production expectations.
Sources
End-to-end realtime acceptance
Run staging scenarios that cover common turns, tool or interruption behavior, retries, and degraded network conditions before switching any production traffic off beta.
Sources
Continue the site
Keep moving through the decision from here.
Use the groups below to move laterally through the decision, not back out into another doc hunt.
Related pages
Stay in the same decision neighborhood instead of backing out to search.
OpenAI deprecations tracker
Provider-specific hub for OpenAI shutdown dates, migration briefs, and replacement paths.
Open pageGPT-4o preview audio and realtime models
Migration brief for dated preview audio and realtime model names moving to current families.
Open pageAssistants API shutdown date
Shutdown-date brief for Assistants API with migration mapping and replacement path.
Open pageCompare pages
Open the pages that turn this topic into a side-by-side decision.
Replacement pages
Use the likely substitutes, migration targets, or fallback choices as the next click.
OpenAI web search pricing
Tool-cost brief for web search pricing across standard and preview search paths.
Open pageOpenAI container pricing
Tool-cost brief for code interpreter container runtime and how it stacks with model spend.
Open pageOpenAI API pricing calculator
Interactive calculator for model tokens, hosted tools, and runtime in one estimate.
Open pageSource category pages
Trace the source families behind this page instead of opening random docs in isolation.