Fixie.ai
Fixie.ai is a B2B platform that enables the building of conversational AI agents. These agents are designed to answer questions, take actio...
Last verified:
What is Fixie.ai?
Fixie.ai (now Ultravox.ai) is a real-time, speech-native voice AI platform that enables developers and companies to build human-like conversational voice AI agents. Unlike traditional Voice AI systems that convert speech to text before processing, Ultravox processes audio directly using its speech-native model, Ultravox v0.7, achieving human-like latencies under 500ms and preserving paralinguistic cues like tone, cadence, and pitch that are lost in transcription.
Key features include developer-friendly REST APIs, powerful SDKs for web and mobile platforms, built-in telephony integrations with major providers (Twilio, Telnyx, Plivo, Exotel), UltraVAD v0.1 neural voice activity detection for intelligent turn-taking, open-weight models for transparency, support for 42 languages, 60% better transcription accuracy than competitors, function calling and tool integration, RAG (retrieval-augmented generation) for domain-specific knowledge, outbound call scheduling, and custom voice cloning capabilities. The platform is designed for building voice agents that follow instructions reliably and interact with third-party systems effectively.
Ultravox is ideal for development teams building voice AI applications, customer service automation companies, telephony providers, startups creating conversational AI products, enterprises scaling voice AI at massive scale, and researchers interested in open-weight speech models. The platform serves anyone needing fast, natural, real-time voice conversations without the latency and robotic feel of cascaded ASR→LLM→TTS pipelines.
Fixie.ai pricing
Pricing model: Free
Free To Start: First 30 minutes of conversation is free, then $0.05 per minute (including TTS) for up to 5 concurrent calls. Pay Go plan: $0/month, perfect for starting out and experimenting, includes 30 min of free calls, $0.05 per minute after, no surge pricing, unlimited playground calls, up to 5 concurrent calls with some hard limits on concurrency. Pro plan: $100/month when billed yearly, recommended for companies starting to scale, includes everything in Pay Go plus no hard caps on concurrency, outbound call scheduler, 5 custom voices, 20 corpora for RAG. Enterprise plan: Custom pricing, designed for massive scale with priority SLA, org support, and customizable everything, includes dedicated infrastructure, private model fine-tuning, SOC2 compliance, and 24/7 technical support.
Fixie.ai pros
- Speech-native model processes audio directly without ASR transcription
- Under 500ms latency for human-like real-time conversations
- 60% better transcription accuracy than GPT-4o Realtime and Gemini
- 18% improvement in speech-based question answering
- Supports 42 languages for global applications
- Open-weight models committed to open science
- Developer-friendly REST APIs with easy integration
- Powerful SDKs for web and mobile platforms
- Built-in telephony integrations with Twilio, Telnyx, Plivo, Exotel
- UltraVAD v0.1 neural VAD for intelligent turn-taking detection
- No hard concurrency caps on Pro plan
- Function calling and tool integration for complex operations
- RAG support with 20 corpora on Pro plan for domain knowledge
- 5 custom voices included on Pro plan
- Outbound call scheduler on Pro plan
- State-of-the-art 91.8% Big Bench Audio score without reasoning
- 97% Big Bench Audio score with thinking enabled
- Handles noisy environments and low-quality microphones effectively
- Cost-effective at approximately 1/3 the cost of GPT-4o Realtime
Fixie.ai cons
- Speech generation feature is coming soon, not yet available
- Free tier limited to 30 minutes of free calls per month
- Pay Go plan has hard limits on concurrency
- Pro plan costs $100/month when billed yearly
- $0.05 per minute after first 30 minutes can add up for high volume
- Enterprise plan requires custom pricing negotiation
- Requires API key and developer knowledge to integrate
- Best suited for teams with engineering resources
- Telephony requires separate provider credentials setup
- Model upgrades may require code changes for optimal use
Frequently asked questions about Fixie.ai
What is Ultravox and how is it different from traditional Voice AI?
Ultravox is a speech-native large language model trained to understand speech directly, just like humans do. Unlike traditional cascaded Voice AI systems that convert speech to text, process it through an LLM, then convert back to speech (which adds latency and loses paralinguistic signals like tone and pitch), Ultravox processes audio directly. This eliminates transcription latency and preserves meaning-influencing signals, resulting in fast, natural conversations under 500ms latency.
What is the Big Bench Audio score and how does Ultravox perform?
Big Bench Audio is a benchmark for evaluating speech understanding in AI models. Ultravox v0.7 achieves a state-of-the-art 91.8% score without reasoning and an industry-leading 97% with thinking enabled, outperforming closed-source models like GPT-4o Audio, GPT-4.1, Claude Sonnet 4.5, Gemini Live, and Nova 2 Preview when latency is factored in.
How much does Ultravox cost and what plans are available?
Ultravox offers four tiers: Free To Start (first 30 minutes free, then $0.05/minute), Pay Go ($0/month with 30 min free calls, $0.05/minute after, up to 5 concurrent calls), Pro ($100/month billed yearly with no concurrency caps, outbound scheduler, 5 custom voices, 20 RAG corpora), and Enterprise (custom pricing for massive scale with priority SLA and dedicated support).
What telephony providers does Ultravox support?
Ultravox has native support for Twilio, Telnyx, Plivo, and Exotel. It also supports voice platforms like Voximplant and jambonz, and provides SIP integration to connect existing phone systems. With built-in telephony integration, you can make outbound calls directly without managing external telephony provider APIs.
What SDKs are available for integrating Ultravox?
Ultravox provides powerful SDKs for every major platform across web and mobile. There are six SDK implementations available, though specific languages aren't listed on the homepage. The SDKs enable real-time voice conversations, function calling, custom agent behaviors, and call control for building production voice AI applications.
Is Ultravox open source?
Yes, Ultravox is built on open-weight models, and the company is committed to open science. They share their research and findings publicly on platforms like Hugging Face and GitHub, hoping their work can help humanity advance. The core model Ultravox v0.7 and previous versions are available as open-weight speech language models.
What languages does Ultravox support?
Ultravox supports 42 languages, an expansion from 15 languages in earlier versions. This expanded language support enables broader global application for voice AI agents across different markets and regions.
What is UltraVAD and how does it work?
UltraVAD v0.1 is Ultravox's neural voice activity detection model that predicts conversation states and turn-taking. It recognizes when a user is likely finished speaking, identifies typical pause patterns, and distinguishes between a thoughtful pause and the end of a turn, enabling more natural conversation flow without awkward interruptions or delays.
Can I build voice agents without coding experience?
The platform offers a no-code interface for prototyping and testing agents through the Ultravox Console where you can create your first voice AI agent in 2 minutes by entering a name and system prompt. However, for production integration and advanced features like telephony, SDK integration, and API usage, developer knowledge is required.
What is the difference between Agents and Direct Calls in Ultravox?
Agents are reusable templates that define assistant behavior, personality, and capabilities - create once, use for multiple calls with consistent experience, version control, and team collaboration. Direct Calls are one-time configurations where you specify all settings for each individual call, best for quick testing and simple one-off use cases. For all new projects, Ultravox strongly recommends starting with Agents for better consistency, reusability, and maintainability.