OpenAI Realtime API Beta Shutdown: Migration Guide & Real-Time Audio Alternatives

Published 2026-04-10 · Not yet reviewed · Figures compiled 2026-04-10, not re-checked since · 1 OpenAI pricing change tracked

0 days
until Realtime API beta shutdown
May 7, 2026 · OpenAI stability: VOLATILE
0
Days Remaining
6
Alternatives Compared
4
With Free Tiers
4
Breaking Changes

What's happening: OpenAI is deprecating the Realtime API beta on May 7, 2026. The beta endpoints (which required the OpenAI-Beta: realtime=v1 header) will stop working. The GA (stable) Realtime API is the replacement.

Easiest migration: Remove the beta header, update session creation to use client_secrets, and add session_type. If you are already using the OpenAI SDK, the changes are minimal. The GA API uses the same WebSocket protocol with updated event names.

Alternatives exist: If you are reconsidering OpenAI for real-time audio, Deepgram ($200 free credit, $0.0043/min), AssemblyAI (free tier), and Google Cloud Speech-to-Text (60 min/month free) offer real-time transcription at lower per-minute costs.

Jump to section

  1. Breaking Changes
  2. Alternative Comparison Table
  3. Pricing Comparison
  4. Migration Paths
  5. Code Migration Examples
  6. FAQ
  7. OpenAI Change Timeline
  8. Recommendations
  9. Methodology

Breaking Changes: Beta to GA

Four key changes required when migrating from the Realtime API beta to the stable GA release.

1. Remove the Beta Header

The OpenAI-Beta: realtime=v1 header is no longer needed. The GA Realtime API is the default. Remove this header from all requests.

2. New Ephemeral Key Endpoint

Use POST /v1/realtime/client_secrets to generate ephemeral keys for client-side WebSocket connections. This replaces the beta session creation flow.

3. Required session_type Parameter

You must now specify session_type when creating sessions: "speech-to-speech" for bidirectional voice conversations or "transcription" for audio-to-text. The beta used a single session type for both.

4. Updated Event Names and Payloads

Some WebSocket event names and payload structures have been updated in the GA release. Review the official documentation for the updated event reference.

Real-Time Audio API Alternatives

All 6 alternatives compared. Migration effort rated from the perspective of an OpenAI Realtime API integration.

Provider Free Tier Pricing Capability Latency Migration
OpenAI Realtime API (GA) Paid only (pay-per-token) $0.06/min audio input, $0.24/min audio output Speech-to-speech, transcription Low (~200ms) Minimal
Deepgram $200 free credit $0.0043/min (Nova-2) Real-time speech-to-text Very low (~100ms) Moderate
AssemblyAI Free tier available $0.0065/15s (~$0.026/min) Real-time transcription, LeMUR Low (~300ms) Moderate
Azure OpenAI Realtime $200 credit (new accounts) Same as OpenAI (enterprise pricing) Speech-to-speech, transcription Low (~200ms) Low
ElevenLabs 10K characters/month free $0.30/1K characters Real-time text-to-speech Very low (~75ms) Moderate
Google Cloud Speech-to-Text 60 min/month free $0.006/15s (~$0.024/min) Real-time speech-to-text Low (~200ms) Moderate
OpenAI vs alternatives: OpenAI Realtime API is unique in offering speech-to-speech (bidirectional voice conversations with an AI model). Most alternatives focus on either speech-to-text (Deepgram, AssemblyAI, Google) or text-to-speech (ElevenLabs). If you need full voice conversation capability, OpenAI GA or Azure OpenAI are your primary options.

Pricing Comparison

Per-minute costs across all providers. OpenAI Realtime beta pricing shown for reference.

Provider Free Tier Per-Minute Cost Features
OpenAI Realtime (beta) None (shutting down) $0.06/min in, $0.24/min out Speech-to-speech + transcription
OpenAI Realtime (GA) None $0.06/min in, $0.24/min out Speech-to-speech + transcription
Deepgram $200 credit $0.0043/min (Nova-2) Speech-to-text, 30+ languages
AssemblyAI Free tier ~$0.026/min Transcription + LeMUR AI
Azure OpenAI Realtime $200 credit Enterprise pricing Same as OpenAI + Azure compliance
ElevenLabs 10K chars/mo $0.30/1K characters Text-to-speech, voice cloning
Google Cloud STT 60 min/mo ~$0.024/min 125+ languages, streaming
Cost comparison: OpenAI Realtime API is significantly more expensive per minute than speech-to-text alternatives because it includes AI model inference (GPT-4o) in the pipeline. If you only need transcription, Deepgram at $0.0043/min is roughly 14x cheaper than OpenAI's audio input rate. However, for full speech-to-speech with AI reasoning, OpenAI remains the most integrated option.

Migration Paths

Three paths depending on your use case. The right choice depends on whether you need speech-to-speech, transcription only, or voice synthesis.

Path 1: Stay with OpenAI (Beta to GA)

The easiest migration. Remove the beta header, update session creation to use /v1/realtime/client_secrets, add session_type, and update any changed event names. Same SDK, same pricing, same capabilities.

Best for: Existing OpenAI Realtime users who need speech-to-speech and want minimal code changes

Path 2: Transcription-Only (Deepgram, AssemblyAI, Google)

If you only need speech-to-text, dedicated transcription services offer better per-minute pricing and often lower latency. Deepgram Nova-2 leads on accuracy and speed. AssemblyAI adds AI-powered analysis via LeMUR. Google offers the widest language support (125+).

Best for: Applications that process audio input but generate text responses, transcription services, meeting recorders

Path 3: Voice Synthesis (ElevenLabs)

If your use case is generating spoken audio from text, ElevenLabs offers the lowest latency (~75ms) and highest quality voice synthesis with voice cloning capabilities. 10K characters/month free to start.

Best for: Voice assistants, audiobook generation, voice cloning, accessibility features

Code Migration Examples

Python: Beta to GA Migration

Key changes to your server-side session creation:

# Before: Beta session creation import openai client = openai.OpenAI() response = client.chat.completions.create( model="gpt-4o-realtime-preview", # Beta required OpenAI-Beta header (set automatically by SDK) extra_headers={"OpenAI-Beta": "realtime=v1"}, ) # After: GA session creation with client_secrets import openai client = openai.OpenAI() # Create ephemeral key for client-side WebSocket response = client.post( "/v1/realtime/client_secrets", body={ "model": "gpt-4o-realtime", "session_type": "speech-to-speech", # NEW: required }, ) ephemeral_key = response["client_secret"]["value"]

Node.js: Beta to GA Migration

Same pattern — update session creation and remove beta header:

// Before: Beta WebSocket connection const ws = new WebSocket( "wss://api.openai.com/v1/realtime?model=gpt-4o-realtime-preview", { headers: { "Authorization": "Bearer " + apiKey, "OpenAI-Beta": "realtime=v1", // REMOVE this }, } ); // After: GA — get ephemeral key, then connect const resp = await fetch("https://api.openai.com/v1/realtime/client_secrets", { method: "POST", headers: { "Authorization": "Bearer " + apiKey, "Content-Type": "application/json", }, body: JSON.stringify({ model: "gpt-4o-realtime", session_type: "speech-to-speech", // NEW: required }), }); const { client_secret } = await resp.json(); const ws = new WebSocket( "wss://api.openai.com/v1/realtime?model=gpt-4o-realtime", { headers: { "Authorization": "Bearer " + client_secret.value } } );

Alternative: Deepgram Real-Time Transcription

For speech-to-text only, Deepgram offers a simpler WebSocket API with lower per-minute costs:

// Deepgram real-time transcription (Node.js) const { createClient, LiveTranscriptionEvents } = require("@deepgram/sdk"); const deepgram = createClient("YOUR_DEEPGRAM_API_KEY"); const connection = deepgram.listen.live({ model: "nova-2", language: "en", smart_format: true, }); connection.on(LiveTranscriptionEvents.Transcript, (data) => { const transcript = data.channel.alternatives[0].transcript; console.log("Transcript:", transcript); }); // Send audio data to connection.send(audioBuffer)
session_type options: The GA Realtime API requires specifying "speech-to-speech" for bidirectional voice conversations (the model speaks back) or "transcription" for audio-to-text only. The beta handled both in a single session type, so you need to choose which mode your application uses.

Frequently Asked Questions

When does the OpenAI Realtime API beta shut down?
OpenAI is deprecating the Realtime API beta on May 7, 2026. After this date, requests using the OpenAI-Beta: realtime=v1 header will stop working. The stable (GA) Realtime API continues to function and is the direct replacement.
What are the key breaking changes from beta to GA?
There are four main changes: (1) Remove the OpenAI-Beta: realtime=v1 header, (2) Use the new POST /v1/realtime/client_secrets endpoint for ephemeral keys instead of the beta session creation flow, (3) Specify session_type as either 'speech-to-speech' or 'transcription' when creating sessions, and (4) Some event names and payload structures have been updated.
Do I need to change my OpenAI SDK version?
If you are using the latest OpenAI Python SDK (1.x+) or Node.js SDK, the GA Realtime API is supported. The main code changes are removing the beta header, updating the session creation flow to use client_secrets, and adding session_type to your configuration. No major SDK upgrade is required.
What are the best alternatives to OpenAI Realtime API?
For speech-to-text: Deepgram (Nova-2 model, $200 free credit, very low latency) and AssemblyAI (free tier, includes AI analysis via LeMUR). For text-to-speech: ElevenLabs (10K characters/month free, ultra-low latency voice synthesis). For enterprise: Azure OpenAI Realtime (same API, Azure compliance). For multi-language: Google Cloud Speech-to-Text (125+ languages, 60 min/month free).
Is the OpenAI Realtime API GA more expensive than the beta?
The GA pricing model is the same as the beta: audio input costs approximately $0.06/minute and audio output costs approximately $0.24/minute (based on token pricing). There is no price increase with the GA release. However, if cost is a concern, alternatives like Deepgram ($0.0043/min) offer significantly lower per-minute pricing for speech-to-text use cases.

OpenAI Change Timeline

Changes tracked in our deal changes database:

Date Change Impact
May 7, 2026 Realtime API beta endpoints deprecated. Developers must remove OpenAI-Beta header, use new client_secrets endpoint, specify session_type, and update event names. GA Realtime API is the direct replacement. HIGH

Recommendations

Best Alternative for Each Use Case

Fastest migration (recommended for most):

OpenAI Realtime API GA — same SDK, same pricing. Remove the beta header, update session creation, add session_type. If it worked in beta, it will work in GA with minimal changes.

Best for transcription:

Deepgram Nova-2 — $200 free credit, $0.0043/min (14x cheaper than OpenAI audio input). Industry-leading accuracy and very low latency (~100ms). Supports 30+ languages.

Best for transcription + AI analysis:

AssemblyAI — real-time transcription plus LeMUR for summarization, sentiment analysis, and Q&A on transcribed content. Free tier available.

Best for enterprise:

Azure OpenAI Realtime — same API as OpenAI with Azure compliance, data residency, and enterprise support. $200 credit for new accounts.

Best for voice synthesis:

ElevenLabs — ultra-low latency (~75ms) text-to-speech with voice cloning. 10K characters/month free. Best quality synthetic voices on the market.

Best for multi-language:

Google Cloud Speech-to-Text — 125+ languages and variants, 60 min/month free. Best choice if you need broad language coverage.

Methodology

How we track this data: AgentDeals monitors free tier changes across 1,580 developer tools in 66 categories. The Realtime API beta deprecation is tracked in our shutdown tracker and stability dashboard.

Migration recommendations: Based on API documentation review, SDK compatibility analysis, and community reports. Pricing data verified against official provider pricing pages as of 2026-04-10. Free tier availability confirmed via official documentation.

For real-time data, use our stability dashboard, Atom feed, or MCP server. Full dataset available via REST API.

Related Guides

Get this data in your AI editor

Track real-time API shutdowns and compare developer tool free tiers from your AI assistant. Get stability ratings, migration alerts, and pricing comparisons — directly in your editor.

claude mcp add agentdeals -- npx -y agentdeals