Top 6 of 22 researched candidates
Alternatives to OpenAI Realtime API
Find a better fit for what you need. Compare alternatives to OpenAI Realtime API with sourced features, strengths and trade-offs side by side.
On this page
Your reason for switching
What needs to be different?
Choose up to three requirements, or see the closest matches now ↓.
Evidence-backed shortlist
Closest researched alternatives to OpenAI Realtime API
Ordered by the requirements met, strengths preserved, explicit trade-offs and research coverage. Sponsorship does not enter the order.
- 01
ElevenLabs Agents
Hosted platform for building voice and chat agents, with telephony, SDKs and a choice of LLM.
From FreeWhy it's on this listFor the reader whose actual constraint is voice quality and cloning, not the model. ElevenLabs' own documentation confirms bring-your-own-LLM across OpenAI, Anthropic, Google Gemini or a custom endpoint, plus 10,000+ voices with instant cloning. The tradeoff: no public self-serve pricing to compare against OpenAI's transparent per-token rate -- ElevenLabs is sales-led.
Where it improves on OpenAI Realtime API
- +Runs autonomously · YesAgents run whole calls and chats unattended, including scheduled outbound campaigns, with explicit system tools to end a call or hand off to a human number.Evidence ↗
- +Claude · YesAnthropic models are natively selectable in agent settings; a few older Claude models drop out in EU data-residency workspaces.Evidence ↗
- +Open models · YesOpen-weight models run two ways: ElevenLabs-hosted Qwen in the native model list, or Llama and Mixtral through a custom LLM endpoint such as Together AI, Groq or SambaNova.Evidence ↗
Trade-offs against OpenAI Realtime API
- −Admin controls · PartialGuardrails constrain what an agent may say and per-tool approval constrains what it may do, but Guardrails is in Alpha and enabling MCP servers is not restricted to admins.Evidence ↗
- −Audit log · PartialOCSF-schema logs over 100 administrative endpoints, pullable into a SIEM, but gated on Enterprise tier plus an API key holding audit_log_read.Evidence ↗
35/35 researched criteria - 02
AudioCodes Live Hub
Self-service cloud platform for building voice AI agents and connecting them to real telephony channels.
From $0.05/minuteWhy it's on this listRuns autonomously · YesVoice AI Agents are designed to handle an entire inbound call end to end, answering questions and completing tasks with no human agent involved.Evidence ↗
Where it improves on OpenAI Realtime API
- +Claude · PartialClaude models are supported only as bring-your-own custom models configured with your own Anthropic API key; the pre-deployed model list carries OpenAI and Google models only.Evidence ↗
- +Open models · PartialOpen-weight models such as gpt-oss and Gemma are reachable only through the bring-your-own-model path with your own Groq, Cerebras or OpenAI keys, not from the pre-deployed set.Evidence ↗
Trade-offs against OpenAI Realtime API
- −Agent permissions · PartialAn agent calls any tool you configured without a per-call approval step, so control is exercised up front by choosing its tools; the MCP management connection does offer an explicit read-only mode.Evidence ↗
33/35 researched criteria · 2 not established - 03
Synthflow
Enterprise voice AI platform with in-house telephony for inbound and outbound call automation.
From FreeWhy it's on this listRuns autonomously · YesAgents answer and place calls unattended, escalating to a human only on triggers you configure.Evidence ↗
Where it improves on OpenAI Realtime API
- +Model choice · PartialYou pick the model per agent, but only from Synthflow's list of GPT and Synthflow LLM options.Evidence ↗
- +In your editor · PartialNo Synthflow editor extension; Claude, Claude Code and Cowork can drive the workspace over MCP from where a developer already works.Evidence ↗
Trade-offs against OpenAI Realtime API
- −On the command line · NoThe deployment channels are enumerated as telephony, WhatsApp, WebSocket, chat, API, Workflows, Zapier and GoHighLevel; Synthflow ships no command-line tool.Evidence ↗
- −Open source · NoSubscriber Terms grant a non-transferable licence to use the software and forbid decompiling it.Evidence ↗
- −Audit log · PartialAPI logs give an audit trail of every request an agent makes, with retention set by plan; no admin-action audit log is documented.Evidence ↗
35/35 researched criteria - 04
Vapi
Developer platform for building, testing and deploying voice AI agents that make and receive phone calls.
From $0.05/minuteWhy it's on this listFor the reader who wants the call-orchestration layer built for them instead of hand-rolling it against a raw model API. Vapi's own site confirms a unified platform for voice, conversation flow, telephony and integrations, with its own pricing page confirming a Build plan at $0.05/minute (60+ minutes included, model costs passed through at provider rates, $10/month per additional concurrent line) and an annual Scale plan for volume. The tradeoff: a per-minute platform fee on top of model costs, versus OpenAI's raw audio-token-only billing.
Where it improves on OpenAI Realtime API
- +Runs autonomously · YesAgents take whole calls end to end and escalate to a human only when the configured path says so.Evidence ↗
- +Claude · YesClaude is selectable as the assistant's LLM, on Vapi's integration or your own Anthropic account.Evidence ↗
- +Open models · YesOpen-weight models reach the pipeline through Groq, Together AI, Mistral, DeepSeek and OpenRouter, or a self-hosted server.Evidence ↗
Trade-offs against OpenAI Realtime API
- −Agent permissions · PartialTools fire mid-call without human approval; a per-tool rejection plan is the mechanism that withholds an action.Evidence ↗
- −Audit log · PartialThe dashboard keeps org-wide API, call and webhook logs and assistant version history; no member-action audit trail is documented.Evidence ↗
- −Opt out of training · PartialThe privacy policy reserves call content for model training and names no toggle; the $1,000/month ZDR add-on is the only documented way to stop it being kept at all.Evidence ↗
35/35 researched criteria - 05
Deepgram Voice Agent API
One WebSocket API that runs speech-to-text, LLM orchestration and text-to-speech for real-time voice agents.
From $0.05/minuteWhy it's on this listFor the reader who wants multi-provider flexibility with compliance-grade deployment options. Deepgram's own site confirms bring-your-own-LLM or TTS while keeping Deepgram's orchestration and streaming pipeline, deployable fully managed, single-tenant, in VPC, or fully self-hosted for HIPAA, GDPR and data-residency needs, at a flat $4.50/hour with the full stack. The tradeoff: a flat hourly rate rather than OpenAI's pure per-token metering.
Where it improves on OpenAI Realtime API
- +Runs autonomously · YesThe API runs the whole listen-think-speak loop live, handling interruptions and calling functions to take action without a human in the conversation.Evidence ↗
- +Claude · YesAnthropic is a Deepgram-managed provider, so no endpoint of your own is needed; Claude Sonnet models bill at the Advanced tier and Claude Haiku at Standard.Evidence ↗
- +Open models · YesOpen-weight LLMs are listed as supported models: gpt-oss-20b through Groq, which requires you to supply the endpoint, and an NVIDIA Nemotron model that Deepgram manages. Deepgram's own speech and voice weights are not distributed.Evidence ↗
Trade-offs against OpenAI Realtime API
- −Multi-agent · PartialHandoff between specialised agents is a documented pattern with a reference implementation you run yourself: your orchestrator opens a fresh Voice Agent session per agent and summarises context between them. The API itself has no built-in agent-to-agent coordination.Evidence ↗
- −Agent permissions · PartialWhat the agent may do is bounded by the function definitions you send in Settings, and you choose whether each function executes client-side or against an endpoint you host. No approval or confirmation gate is documented: once a function is defined, the model may call it mid-conversation.Evidence ↗
- −Admin controls · PartialRole-based control exists over the account, not over the agent: owner, admin and member roles plus scoped API keys govern who may read usage, create keys or change billing. What the agent itself may say or call is bounded by your prompt and function definitions, with no admin-side policy engine documented.Evidence ↗
35/35 researched criteria - 06
Cartesia
Real-time speech models and a managed runtime for production phone and web voice agents.
From FreeWhy it's on this listRuns autonomously · YesA deployed agent runs a whole call unattended, handling turn-taking, interruptions and tool calls; it transfers to a human only when a configured condition matches.Evidence ↗
Where it improves on OpenAI Realtime API
- +Claude · YesAnthropic Claude models are named in the SDK's supported-models table and used in its examples.Evidence ↗
- +Open models · PartialThe LLM layer routes through LiteLLM to 100+ providers, but the vendor names only Anthropic, OpenAI and Google, and its own speech models are API-only.Evidence ↗
Trade-offs against OpenAI Realtime API
- −Multi-agent · NoSystem tools enumerate three slots: end call, DTMF, and transfer to an E.164 phone number. An agent therefore hands off to a number, not to another Cartesia agent.Evidence ↗
- −Agent permissions · PartialLimits are prompt-level guardrails plus per-tool execution settings (timeout, immediate vs async, speak-before-acting); there is no separate policy engine.Evidence ↗
- −Admin controls · PartialRoles are Admin and Member only: an admin controls the org, invitations and members, not what an agent is permitted to do at runtime.Evidence ↗
35/35 researched criteria
Full datasheet
Compare OpenAI Realtime API with the shortlist
Replace any candidate, show every criterion, or keep only differences and unknowns.
Compare alternatives to OpenAI Realtime API
Keep OpenAI Realtime API as the baseline and choose up to three alternatives. Differences and unanswered criteria appear first.
The digest
Done comparing? Watch the field move. Which agents actually ship, one short email, every Tuesday.
Advertisement
Sponsor this page. Alternatives pages are read by buyers actively choosing an agent, first-party inventory, clearly labelled, sold by the week.
Sponsor this page →Featured · top of your category →Method
How the shortlist is calculated
Comparable first. We consider listings in the same category as OpenAI Realtime API and alternatives previously researched for this page.
Requirements first. We prioritise your selected needs, then shared strengths, fewer documented trade-offs, more complete research and finally alphabetical order.
Unknown remains unknown. Missing evidence is never converted into a benefit or a trade-off. Every improvement shown above comes from the same sourced criterion as the listing.
Independent order. Payment does not affect selection or ranking. Paid placements are labelled Advertisement.
Where next? Back to the full research, deeper into the category, or get your own agent into the index.
Advertise here
Reach buyers mid-decision. Reach builders choosing their next agent. Promote your brand with a display placement or bring your listing into focus with Featured.
Explore owner options →Advertise on this page →The digestFree
Which agents actually ship.
What we re-checked, what got added, and one number from the index. Tuesdays.
One-click unsubscribe