OpenAI Realtime API

https://platform.openai.com/docs/guides/realtime

OpenAI's Realtime API for native speech-to-speech multimodal interaction with GPT-4o — launched Oct 2024 (DevDay). Skips the STT → LLM → TTS pipeline by passing speech directly to GPT-4o. The platform substrate that Vapi/Retell/Bland integrate as one of their LLM choices.

At a glance

Type
Speech-to-speech LLM API (OpenAI Realtime)
Tier
T1
Created
2024-10
Latest release
not applicable — not OSS
License
not applicable — not OSS
Pricing
Audio token-based: $40/$80 per 1M input/output audio tokens (Oct-2024 launch)
Funding
OpenAI parent — $40B round Mar-2025 ($300B val)

Taxonomy

storage
kv
retrieval
injection
persistence
session
update
agent-controlled
unit
turn
governance
opaque
conflict
stateless

When to use

Optimised for: Native speech-to-speech with GPT-4o; lowest-latency LLM voice tier

Anti-fit: Stateless platform — caller builds orchestration, memory, telephony adapter

Pros & cons

Pros

Native speech-to-speech (skips STT/TTS); OpenAI brand + price/perf; WebRTC support.

Cons

Audio token pricing is expensive at scale; stateless (caller builds memory/orchestration); OpenAI-only model choice.

Claims & capabilities

Launched DevDay Oct-2024; GPT-4o speech-to-speech native; WebSocket + WebRTC support

Technical surface

API surface
WebSocket + WebRTC + REST
Backend storage
Stateless session API; caller manages durability
Deployment
API (Cloud); no self-host
Embedding model
not applicable — not a memory product
Multi-tenancy
API multi-tenant
MCP
searched not found
A2A
searched not found
OpenTelemetry
searched not found

Similar systems

Other voice agent platforms in the catalog, ranked by inbound references.

  • Cartesia Sonic T1

    Cartesia's Sonic TTS — state-space-model-based voice generation, lowest published TTS latency (~40ms). Cartesia raised $64M Series A Q1 2025 (Index + Lightspeed); founded by Mamba authors. Agent platform layer ('Cartesia Line') wraps Sonic with telephony + LLM.

  • Vapi T1

    Voice-AI developer platform — STT + LLM + TTS pipelines wired to telephony (Twilio / SIP). Used to build outbound + inbound phone agents at scale. Raised $20M Series A from Bessemer + Y Combinator (Dec 2024) at $130M valuation; revenue reported >$10M ARR by Q1 2025.

  • Bland AI T1

    Voice-AI platform with custom STT + LLM + TTS stack ('the whole pipeline'). Raised $40M Series A Apr-2024 (Lightspeed + Anthology). Distinguished by 'pathway' graphs (deterministic call flows) and aggressive outbound-calling positioning.

  • CallRail Conversation Intelligence T2

    CallRail's AI-driven conversation intelligence layer atop their call-tracking product. Marketing-vertical voice analytics — auto-tagging, sentiment, lead scoring. CallRail is profitable / private (Sageview-backed); not a voice-agent runtime but the BI / analytics adjacent layer.

  • ElevenLabs Conversational AI T1

    ElevenLabs launched Conversational AI Nov-2024 — STT + LLM + their own market-leading TTS in one pipeline. Combines vendor's voice cloning + multilingual TTS strengths with hosted call orchestration. ElevenLabs raised $180M Series C Jan-2025 at $3.3B val.

  • Hume EVI T1

    Hume's Empathic Voice Interface (EVI) — voice agent platform with vocal-affect understanding. Hume raised $50M Series B Mar-2024 (EQT). Distinguished from pure LLM voice agents by emotion/prosody modelling; uses Hume's proprietary speech-prosody models.

Row last verified 2026-05-14. Catalog data is CC-BY-4.0 — see how to read this.