OpenAI Realtime API (gpt-realtime)

Best for: Regulated or enterprise workloads - compliance attestations and an enterprise plan; AI agents and automation - an agent-ready surface (MCP / llms.txt); Teams needing broad API coverage out of the box

Avoid if: You want to try it free before paying

Pricing model

Usage-based [2]developers.openai.com/api/docs/pricing“gpt-realtime-2 Audio Input: $32.00 per 1M tokens; Audio Output: $64.00 per 1M tokens; gpt-realtime-translate: $0.034 / minute; gpt-realtime-whisper: $0.017 / minute”developers.openai.com/api/docs/models/gpt-realtime-2“Audio Input: $32.00; Cached input: $0.40; Audio Output: $64.00; Text Input: $4.00; Text Output: $24.00 per 1M tokens”

Published pricing

✓ Yes [3]developers.openai.com/api/docs/pricing“gpt-realtime-2 Audio Input: $32.00 per 1M tokens; Audio Output: $64.00 per 1M tokens; Cached Audio Input: $0.40 per 1M tokens”

Free tier

✗ No

Self-serve signup

✓ Yes

Requires sales call

✗ No

Enterprise plan

✓ Yes [4]openai.com/api-scale-tier/“OpenAI's Scale Tier traffic offers a 99.9% uptime SLA and prioritized compute for customers on enterprise agreements.”

• Multilingual

Supported actions

realtime_conversation, speech_to_speech, inbound_telephony, sip_trunking, call_transfer, function_calling, tool_calling, interruption_handling, voice_activity_detection, live_transcription, realtime_translation, mcp_server_integration, websocket_streaming, webrtc_streaming, call_recording, webhook_events, reasoning_effort_control [5]developers.openai.com/api/docs/guides/realtime“SIP: Use for telephony voice agents. WebRTC: Use for browser and mobile clients that capture or play audio directly. WebSocket: Use when your server already receives raw audio from a media pipeline.”developers.openai.com/api/docs/guides/realtime-sip“The Realtime API supports inbound phone calls through SIP (Session Initiation Protocol). A SIP trunking provider converts telephone calls to IP traffic, which OpenAI routes to the Realtime API via a webhook system.”

Regions

North Europe, South Central US, East US 2, West US, EU Data Residency [6]developers.openai.com/api/docs/guides/realtime-sip“For allowlisting OpenAI SIP traffic, dedicated IP ranges are provided for: northeurope: 13.79.45.80/28, southcentralus: 23.98.140.64/28, eastus2: 40.67.149.176/28, westus: 40.83.204.240/28”

Languages

English, Mandarin, Spanish, French, German, Japanese, Korean, Portuguese, Arabic, Hindi, Russian, Italian, Indonesian, 70+ additional input languages via Whisper-based transcription (gpt-realtime-whisper), 70+ input languages for translation via gpt-realtime-translate (output to 13 languages listed above) [7]developers.openai.com/api/docs/models/gpt-realtime-translate“a streaming speech-to-speech translation model for live multilingual audio experiences”community.openai.com/t/new-realtime-voice-models-in-the-api/1380471“GPT-Realtime-Translate translates speech from 70+ input languages into 13 output languages while keeping pace with the speaker.”

Input types

audio stream (WebRTC), audio stream (WebSocket), SIP phone call, text [8]developers.openai.com/api/docs/guides/realtime“WebRTC: Use for browser and mobile clients that capture or play audio directly. WebSocket: Use when your server already receives raw audio from a media pipeline. SIP: Use for telephony voice agents.”

Output types

audio stream, text transcript, tool call results, webhook events, session events

Webhooks

✓ Yes [9]developers.openai.com/api/docs/guides/realtime-sip“Receive realtime.call.incoming webhook events; Accept or reject calls using dedicated API endpoints”

Sandbox / test mode

✗ No

SDK languages

Python, Node.js, .NET, Java, Go, Ruby, Python (Agents SDK), Node.js (Agents SDK) [10]developers.openai.com/api/docs/libraries“OpenAI provides official SDKs for JavaScript/TypeScript, Python, .NET/C# (developed in collaboration with Microsoft), Java (beta), Go (beta), and Ruby.”

MCP server

✓ Yes [11]developers.openai.com/api/docs/guides/realtime“Realtime with tools guide covers Connect function tools, MCP servers, and connectors to a Realtime session.”

SOC 2

SOC 2 Type II [12]trust.openai.com“SOC 2 Type 2: Our products are covered in our SOC 2 Type 2 report and have been evaluated by an independent third-party auditor.”

HIPAA

✓ Yes [13]help.openai.com/en/articles/8660679-how-can-i-get-a-business-associate-agreement-baa-with-openai“OpenAI is able to sign Business Associate Agreements (BAA) in support of customers' compliance with the Health Insurance Portability and Accountability Act (HIPAA).”

GDPR

✓ Yes [14]trust.openai.com“GDPR: Listed as certified”

ISO 27001

✓ Yes [15]trust.openai.com“ISO/IEC 27001:2022: Certified for API, ChatGPT Enterprise, and ChatGPT Edu; ISO/IEC 27017:2015, ISO/IEC 27018:2019, ISO/IEC 27701:2019”

PCI DSS

✓ Yes [16]trust.openai.com“PCI DSS v4.0.1: Covers payment processing components”

Published SLA

✓ Yes [17]openai.com/api-scale-tier/“OpenAI's Scale Tier traffic offers a 99.9% uptime SLA and prioritized compute. Priority processing SLAs are treated the same as Scale Tier SLAs, with service credits offered should they fail to meet those SLAs for customers on enterprise agreements.”

Rate limits

gpt-realtime-2: Tier 1–5 scaling from 200–20,000 RPM and 40,000–15,000,000 TPM; audio billed at 1 token per 100ms input, 1 token per 50ms output. gpt-realtime-translate: 50–850 audio-minutes per minute depending on tier. [18]developers.openai.com/api/docs/models/gpt-realtime-translate“Rate Limits vary by tier, ranging from 50 to 850 minutes of audio per minute depending on your usage tier.”developers.openai.com/api/docs/models/gpt-realtime-2“Tier 1: 200 RPM / 40,000 TPM; Tier 2: 400 RPM / 200,000 TPM; Tier 3: 5,000 RPM / 800,000 TPM; Tier 4: 10,000 RPM / 4,000,000 TPM; Tier 5: 20,000 RPM / 15,000,000 TPM”

Known restrictions

SIP support is inbound calls only (no outbound SIP calling), SIP requires a third-party SIP trunking provider (e.g. Twilio) - telephony is BYO carrier, not included, gpt-realtime-translate and gpt-realtime-whisper support only translation/transcription session types, not full voice-agent conversations, No BYO-LLM or BYO-voice support - only OpenAI models available, No sandbox/test environment - live API keys required from the start, SLA (99.9% uptime) is only available to Scale Tier / enterprise customers, not standard pay-as-you-go users, Beta interface deprecated September 15, 2025; GA interface requires updated headers and event shapes [19]openai.com/api-scale-tier/“OpenAI's Scale Tier traffic offers a 99.9% uptime SLA and prioritized compute for customers on enterprise agreements.”developers.openai.com/api/docs/guides/realtime-sip“Currently supported: Inbound calls only. A SIP trunking provider converts telephone calls to IP traffic.”

• Docs
• API reference

Docs rendering: static · markdown variants served

API style

websocket

Base URL

wss://api.openai.com/v1/realtime

Version

gpt-realtime-2025-08-28

Versioning

url

Stability

ga

Auth methods

api_key

Idempotency keys

✗ No

Error format

vendor-specific

SDKs

• Python openai · repo
• Node.js openai · repo
• .NET OpenAI · repo
• Java openai-java · repo
• Go openai-go · repo
• Ruby openai · repo
• Python (Agents SDK) openai-agents · repo
• Node.js (Agents SDK) openai-agents-js · repo

• Auth docs
• Quickstart
• MCP server
• Deprecation policy

Launched

2024-10-01

GA

2025-08-28

Notable customers

Perplexity, Healthify, Speak, Zillow, Genspark

• ElevenLabs Conversational AI (ElevenAgents)

  "Deploy human-like Conversational AI in minutes. ElevenLabs delivers low latency interactions in dozens of languages with enterprise-grade security."

  Hybrid · free tier · public pricing · self-serve

• Telnyx Voice AI Agents

  "Carrier-grade Voice AI from Telnyx. Sub-200ms latency, 80+ languages, A-level STIR/SHAKEN attestation, and one platform that replaces 5 vendors."

  Usage · public pricing · self-serve

• Twilio ConversationRelay

  "Twilio's Conversation Relay empowers you to build powerful AI voice experiences for your customers. Let Twilio handle the heavy lifting of speech recognition, text-to-speech, and voice synthesis."

  Usage · public pricing · self-serve

• Vapi

  "Build and deploy voice agents that deliver the outcomes you want at the scale your customers need."

  Usage · free tier · public pricing · self-serve

• Cartesia Line

  "Turn any text agent into a world-class conversational agent, deployed anywhere."

  Hybrid · free tier · public pricing · self-serve

• Pipecat Cloud (Daily)

  "The fastest path to production voice AI" - a managed hosting platform for deploying and scaling Pipecat agents in production with built-in infrastructure.

  Usage · public pricing · self-serve

OpenAI Realtime API (gpt-realtime) alternatives · OpenAI Realtime API (gpt-realtime) vs ElevenLabs Conversational AI (ElevenAgents) · All Realtime Voice Agent APIs APIs

Each field above carries a numbered source - hover for a preview, click to jump here.

1. ↑Description: developers.openai.com
2. ↑Pricing model: developers.openai.com · developers.openai.com
3. ↑Published pricing: developers.openai.com
4. ↑Enterprise plan: openai.com
5. ↑Supported actions: developers.openai.com · developers.openai.com
6. ↑Regions: developers.openai.com
7. ↑Languages: developers.openai.com · community.openai.com
8. ↑Input types: developers.openai.com
9. ↑Webhooks: developers.openai.com
10. ↑SDK languages: developers.openai.com
11. ↑MCP server: developers.openai.com
12. ↑SOC 2: trust.openai.com
13. ↑HIPAA: help.openai.com
14. ↑GDPR: trust.openai.com
15. ↑ISO 27001: trust.openai.com
16. ↑PCI DSS: trust.openai.com
17. ↑Published SLA: openai.com
18. ↑Rate limits: developers.openai.com · developers.openai.com
19. ↑Known restrictions: openai.com · developers.openai.com

Every field change, who made it, and when - from our audited data pipeline and editors.

1. 2026-06-21 Capabilities: {} → {"multilingual":true}
2. 2026-06-21 Summary Md: (none) → OpenAI Realtime API is a WebSocket-based service for low-latency, bidirectional…
3. 2026-06-21 Avoid If: (none) → You want to try it free before paying
4. 2026-06-21 Scoring Methodology: (none) → Scores are computed deterministically from this profile's published, sourced fi…
5. 2026-06-21 Score Pricing Transparency: (none) → 85
6. 2026-06-21 Score Setup Speed: (none) → 60
7. 2026-06-21 Score Docs Quality: (none) → 50
8. 2026-06-21 Score Agent Friendliness: (none) → 50
9. 2026-06-21 Score Procurement Friction: (none) → 85
10. 2026-06-21 Score Trust Readiness: (none) → 100
11. 2026-06-21 Best For: (none) → Regulated or enterprise workloads - compliance attestations and an enterprise p…
12. 2026-06-21 Llms Txt Present: (none) → No
13. 2026-06-21 Rendering: (none) → static
14. 2026-06-21 Has Structured Data: (none) → No
15. 2026-06-21 Robots Allows Agents: (none) → Yes
16. 2026-06-21 API Reference URL: (none) → https://platform.openai.com/api/reference/overview
17. 2026-06-21 Markdown Docs Served: (none) → Yes
18. 2026-06-21 Markdown Docs URL: (none) → https://platform.openai.com/docs/guides/realtime.md
19. 2026-06-21 Docs URL: (none) → https://developers.openai.com/api/docs
20. 2026-06-21 MCP Server Available: set to Yes
21. 2026-06-21 Pricing Model: set to usage_based
22. 2026-06-21 Has Published Pricing: set to Yes
23. 2026-06-21 Free Tier Available: set to No
24. 2026-06-21 Self Serve Signup: set to Yes
25. 2026-06-21 Requires Sales Call: set to No
26. 2026-06-21 Enterprise Plan Available: set to Yes
27. 2026-06-21 SOC 2: set to type_2
28. 2026-06-21 HIPAA: set to Yes
29. 2026-06-21 GDPR: set to Yes
30. 2026-06-21 ISO 27001: set to Yes
31. 2026-06-21 PCI DSS: set to Yes
32. 2026-06-21 SLA Published: set to Yes
33. 2026-06-21 SLA URL: set to https://openai.com/api-scale-tier/
34. 2026-06-21 Data Retention Policy URL: set to https://developers.openai.com/api/docs/guides/your-data
35. 2026-06-21 Documented Rate Limits: set to gpt-realtime-2: Tier 1–5 scaling from 200–20,000 RPM and 40,000–15,000,000 TPM;…
36. 2026-06-21 Known Restrictions: set to SIP support is inbound calls only (no outbound SIP calling), SIP requires a thi…
37. 2026-06-21 Auth Methods: set to api_key
38. 2026-06-21 API Version: set to gpt-realtime-2025-08-28
39. 2026-06-21 Versioning Scheme: set to url
40. 2026-06-21 Stability: set to ga
41. 2026-06-21 Deprecation Policy URL: set to https://developers.openai.com/api/docs/deprecations
42. 2026-06-21 MCP URL: set to https://developers.openai.com/mcp
43. 2026-06-21 Quickstart URL: set to https://developers.openai.com/api/docs/guides/voice-agents
44. 2026-06-21 Idempotency Supported: set to No
45. 2026-06-21 Error Format: set to vendor-specific
46. 2026-06-21 Requires Verification: set to No
47. 2026-06-21 Starting Price Usd: set to 0.0192
48. 2026-06-21 Price Basis: set to minute (audio input)
49. 2026-06-21 Launched At: set to 2024-10-01
50. 2026-06-21 GA Date: set to 2025-08-28

This profile is crowd-editable - agents and humans can leave a review or propose a correction with a simple API call. No auth; requests are rate-limited and every submission is reviewed before it goes live. For a field edit, use any key from the Agent JSON in place of FIELD, and include a citation.

curl -X POST https://apio.sh/api/feedback/openai-realtime \
  -H 'Content-Type: application/json' \
  -d '{"kind":"review","rating":5,"body":"Your experience with this API…"}'

curl -X POST https://apio.sh/api/suggest/openai-realtime/FIELD \
  -H 'Content-Type: application/json' \
  -d '{"value":"corrected value","citations":[{"url":"https://source.example/page","excerpt":"supporting quote"}],"note":"what changed and why"}'

All the ways to contribute →