Ultravox

What Ultravox does

Ultravox, operated by Fixie.ai Inc. and headquartered in Seattle, Washington, is a research lab and product company building speech-native voice AI infrastructure for developers and enterprise voice agent teams. Its core technology is the Ultravox Realtime platform, powered by the proprietary Ultravox v0.7 model built on a GLM 4.6 backbone (355B parameters, 160 experts per layer). Unlike cascaded voice stacks that transcribe speech to text before LLM inference, Ultravox performs reasoning directly on audio signals, preserving paralinguistic cues such as tone, cadence, and pitch while delivering approximately 900ms end-to-end latency. The platform also includes the UltraVAD v0.1 neural endpointing model for turn-taking detection and exposes a dedicated inference stack the company manages end-to-end.

The product is delivered through REST APIs, WebSockets, and SDKs spanning JavaScript/TypeScript, Python, React Native, Swift, Kotlin, and Flutter, with native telephony integrations for Twilio, Voximplant, Telnyx, Plivo, and jambonz, and voice output options including Inworld, ElevenLabs, Cartesia, LMNT, Google, and Respeecher. Pricing is anchored on a freemium + usage-based model at $0.05 per minute (30 free minutes monthly), a $100/month Pro subscription that lifts concurrency caps and adds voice cloning and RAG corpora, and custom Enterprise contracts for large-scale deployments with priority SLA.

Ultravox targets developers building voice agents for customer service, lead qualification, and appointment scheduling, as well as enterprise buyers in verticals such as healthcare and e-commerce. Distribution combines product-led growth (self-serve sign-up, free tier, Discord community, GitHub SDKs, Hugging Face open weights) with a sales-led enterprise motion via Request a Demo. The company has raised approximately $17 million across a 2022 seed and a February 2023 Series A led by Redpoint, grew its customer base 18x and its traffic 38x in 2025, and operates with a team of approximately 11 people.

Name

Ultravox

Legal name

Fixie.ai Inc.

Website

https://ultravox.ai

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

11–50 employees

Short description

Ultravox (operated by Fixie.ai Inc.) provides the Ultravox Realtime platform, a speech-native voice AI infrastructure that enables developers and enterprises to build real-time conversational agents via direct audio inference without transcription.

Ownership category

akta.pro rank

Product category

Voice AI Platform

NAICS

Computer Systems Design and Related Services (54151), Custom Computer Programming Services (541511), Translation and Interpretation Services (541930)

SIC

Services-Computer Programming Services (7371), Services-Computer Integrated Systems Design (7373), Services-Telephone Interconnect Systems (7385), Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

Voice Assistants & Voice UI (VUI) (HDAAAFAD)

Keywords

• Speech-native voice AI
• Real-time voice agents
• Voice AI platform
• Speech-to-speech models
• AI agent APIs

Headquarters

HQ city

Seattle

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel, Operations, Marketing or Sales

Revenue model

1. Usage-based Voice AI (Pay As You Go): Primary revenue stream based on per-minute usage of voice AI calls. Includes 30 free minutes per month, then $0.05 per minute for speech processing and TTS. No hard caps on total usage but limited to 5 concurrent calls on free tier.
2. Pro Subscription: Monthly subscription at $100/month (billed annually) providing unlimited playground calls, no hard caps on concurrency, outbound call scheduler, 5 custom voice clones, and 20 corpora for RAG. Includes $0.05/min usage at the same rate as Pay As You Go.
3. Enterprise Custom: Custom pricing for massive scale deployments with priority SLA, organizational support, and fully customizable platform features. Sold via sales-led motion through Request a Demo.
4. SIP Telephony: Additional usage-based revenue from SIP call handling at 0.5¢/minute (Pay As You Go) or 0.48¢/minute (Pro). Bridge transfers incur additional costs as SIP infrastructure remains active.
5. Threads (Token-based): Charges for token usage in thread-based conversations: $2.00 per 1M input tokens (uncached), $15.00 per 1M output tokens. Cached input tokens are free.

Pricing tiers

Go-to-market motion3 records

Distribution channels6 records

Marketing channels7 records

Core offering

Ultravox operates the Ultravox Realtime platform, a speech-native voice AI infrastructure layer that processes audio directly without transcription, enabling developers and enterprises to build real-time conversational voice agents. The platform exposes REST APIs, WebSockets, and SDKs across web and mobile (JavaScript, Python, React Native, Swift, Kotlin, Flutter), integrates with major telephony providers, supports retrieval-augmented generation via Ultravox Corpus, and offers voice cloning with multiple TTS providers. The proprietary Ultravox v0.7 model powers the platform, delivering ~900ms end-to-end latency and state-of-the-art benchmark performance.

Product overview

Ultravox is a research lab and product company offering the Ultravox Realtime platform — a speech-native voice AI infrastructure layer built on the Ultravox v0.7 model (powered by GLM 4.6). Unlike traditional voice AI systems that transcribe speech to text before LLM processing, Ultravox performs inference directly on audio, preserving paralinguistic cues and delivering ~900ms end-to-end latency. The platform is accessed via REST API, WebSockets, and SDKs (JavaScript, Python, React Native, Swift, Kotlin, Flutter). It integrates telephony providers (Twilio, Voximplant, Telnyx, Plivo, jambonz) and voice providers (Inworld, ElevenLabs, Cartesia, LMNT, Google, Respeecher). The Ultravox Corpus service enables RAG-based knowledge retrieval. Voice cloning is available for custom synthetic voices. Pricing spans Pay As You Go ($0.05/min), Pro ($100/month), and Enterprise (custom).

Differentiator

Problem solved

Functional benefit

Brands

• Ultravox Realtime Platform: Developer-friendly real-time voice AI platform with APIs, SDKs, and agentic-ready primitives for building conversational voice AI experiences.
• Ultravox v0.7
• UltraVox VAD v0.1

Products and services

• Ultravox Realtime Platform The core voice AI platform enabling developers and enterprises to build real-time speech-native voice agents. Provides a unified stack combining speech understanding, reasoning, and speech generation without transcription, delivered with ~900ms end-to-end latency.
• Ultravox v0.7 Speech Model State-of-the-art speech-native model using GLM 4.6 (355B parameters, 160 experts per layer) as its LLM backbone. Achieves 91.8% on Big Bench Audio without reasoning and 97% with thinking enabled; performs inference directly on audio without transcription. Released as open weights on Hugging Face.
• Ultravox Voice API REST API underlying the Ultravox Realtime platform with endpoints for accounts, agents, calls, voices, corpora, tools, webhooks, SIP configuration, and telephony credentials. Supports WebSocket-based real-time streaming and webhook events for call.started and call.ended.
• Ultravox Client SDKs Client software development kits enabling developers to integrate Ultravox voice AI into web and mobile applications. Available for JavaScript/TypeScript, Python, React Native, Swift (iOS), Kotlin (Android), and Flutter.
• Ultravox Realtime Telephony Telephony module enabling voice agents to make and receive phone calls via SIP trunking, with built-in integrations for Twilio, Voximplant, Telnyx, Plivo, and jambonz. Supports inbound/outbound calls, SIP call transfers (cold and warm), IVR flow handling, and outbound call scheduling.
• Ultravox Corpus (RAG) Retrieval-augmented generation knowledge base service integrated into Ultravox Realtime. Users upload documents, websites, or other content to create corpora that voice agents query during conversations, with isolation between customers' knowledge bases. Up to 20 corpora on the Pro plan.

Quantifiable outcome

• 91.8% on Big Bench Audio (97% with thinking) - highest score among speech models
• +4 more outcomes

Named customers1 record

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration8 records

AI capability7 records

Feature8 records

Partnerships

Five partnerships are on record, tiered core.

• InworldcoreTechnology or Integration · 21 January 2026Integration of Inworld TTS 1.5 voices into Ultravox Realtime platform. Users can build voice agents using Inworld's pre-built voices in 15 languages or custom voice clones. Joint promotional credit offer of up to $100 for customers using Inworld voices (January 21 - February 28, 2026).
• PipecatcoreTechnology or Integration · 16 January 2026Ultravox Realtime available as speech-to-speech service in Pipecat voice AI framework. Ultravox v0.7 demonstrated superior performance in Pipecat team's independent benchmark, outperforming all other speech-to-speech models tested. Designed as drop-in replacement.
• VoximplantcoreTechnology or IntegrationNative integration enabling SIP calling through Voximplant platform. Voximplant provides hosted telephony platform that connects to Ultravox via WebSockets, managing connections with existing SIP telephony infrastructure. Includes ready-to-use scenarios for incoming and outgoing calls.
• TwiliocoreTechnology or IntegrationTelephony integration allowing AI agents to handle calls via Twilio. Built-in coldTransfer and warmTransfer tools work with Twilio for call handoffs to human operators. Complete working examples available in ultravox-examples GitHub repo.
• Hugging FacecoreTechnology or IntegrationOpen weight model releases published on Hugging Face. Model weights for Ultravox v0.7 (GLM 4.6 backbone) and previous versions available for download. Supports open science philosophy and community research.

Scale indicators9 records

Recent moves5 records

Expansion highlights7 records

Direct peers

• Retell AI: Retell AI provides a voice agent platform with low-latency conversational AI for enterprise customer service and sales. It targets an overlapping enterprise segment with similar telephony integration (Twilio, etc.) and developer-first API access, competing for the same voice AI agent budget.
• AssemblyAI: AssemblyAI provides speech recognition and audio intelligence APIs with a Universal-1 model and emerging real-time/agentic features. It targets the same developer customer as Ultravox with usage-based pricing and a model-plus-API stack, competing for voice AI infrastructure spend.
• Vapi: Vapi is a developer platform for building, testing, and deploying voice AI agents with telephony integrations. It targets an overlapping developer/enterprise customer base as Ultravox and competes in the same voice agent infrastructure layer, though Vapi leans orchestration while Ultravox leans speech-native model.
• Deepgram: Deepgram provides speech-to-text and increasingly speech-to-speech voice AI APIs for developers. It targets the same developer-first, usage-priced customer base as Ultravox, with similar SDK and enterprise motions, though Deepgram has historically focused on STT accuracy rather than speech-native reasoning.
• ElevenLabs: ElevenLabs is a leading voice AI platform spanning TTS, voice cloning, and conversational agents. It is listed as one of Ultravox's supported TTS providers (Inworld, ElevenLabs, Cartesia, etc.) and competes head-on in the developer/enterprise voice agent space, particularly for high-quality voice output.

Broad incumbents

• OpenAI (Realtime API): OpenAI's Realtime API offers speech-to-speech conversations via GPT-4o Realtime. It is a direct competitor to Ultravox in low-latency voice agents, with substantially more resources and bundled distribution, but Ultravox differentiates on accuracy (97.7% vs 86.7% on AIEWF) and latency (0.864s vs 1.536s).
• Google (Gemini Live / Vertex AI Speech): Google's Gemini Live provides a multimodal live voice experience and competes directly with Ultravox's real-time speech-to-speech platform. Ultravox outperforms Gemini Live on AIEWF (97.7% vs 86.0%) and latency (0.864s vs 2.624s), but Google has unmatched cloud distribution.

Emerging players

• Hume AI: Hume AI offers an emotionally intelligent voice AI API (Empathic Voice Interface) that captures emotional and paralinguistic signals. It competes with Ultravox on the dimension Ultravox emphasizes (non-transcript-based, emotion-aware voice), and targets similar enterprise conversational AI buyers.
• Cartesia: Cartesia builds real-time speech models with a focus on state-space model (SSM) architecture for low-latency voice AI. It is listed as a supported TTS provider on Ultravox and overlaps as an emerging speech-native model competitor targeting similar enterprise/developer use cases.

Others

• Pipecat (Daily): Pipecat is an open-source voice AI orchestration framework that integrates Ultravox as a drop-in speech-to-speech service. While not a direct competitor, Pipecat sits in the same developer stack and its independent benchmark showcasing Ultravox v0.7 makes it a key ecosystem partner that drives developer adoption of Ultravox over GPT Realtime or Gemini Live.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/ultravox_dot_ai
• GitHubgithub.com/fixie-ai

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles7 records

Funding overview

Funding rounds3 records

Investors6 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Ultravox