Pomegra
Inworld has acquired Ultravox, the real-time voice-agent platform, on undisclosed terms, adding a developer product and a speech-to-speech research team to its text-to-speech models.
Key Takeaways
- Inworld announced the Ultravox acquisition on October 1, 2026; financial terms were not disclosed.
- Ultravox customers' built-in Inworld voices moved to Realtime TTS-2 at no added cost.
- Ultravox is the former Fixie, which raised a $17 million seed round led by Redpoint Ventures in 2023.
Lead
Inworld, a Mountain View, California AI research lab and inference provider, announced on October 1, 2026 that it has acquired Ultravox, a platform developers use to build real-time voice agents. Neither company disclosed a price, a structure or a valuation. Inworld said members of the Ultravox team have joined the company and will keep building the platform.
The deal is a strategic purchase of a product with an existing developer base, not a pure acquihire. Ultravox keeps operating, and the first integration is already live.
What Does Inworld Get With Ultravox?
Inworld gets a working voice-agent platform and the team behind its turn-taking and interruption handling. Those are the parts of a voice agent that decide when to speak, when to stop and how to respond when a user cuts in. Inworld said the acquisition accelerates its speech-to-speech research and voice agent work.
Ultravox processes speech directly rather than converting it to text first. It supports phone calls, web apps, native apps and WebSockets. Its model is open-weight and MIT-licensed, built on a Llama 3.3 70B backbone with a fine-tuned Whisper large-v3-turbo audio encoder. Pricing is usage-based at $0.05 per minute, with a $100-per-month Pro plan.
Inworld's own pitch is a full stack for consumer AI: speech models, served language models and the realtime inference behind modular APIs. Ultravox adds the orchestration layer on top of that.
Why Did Inworld Buy Rather Than Build?
Inworld bought because turn-taking is hard to get right and Ultravox had already shipped it. Inworld CEO Kylan Gibbs said Ultravox "has built a platform that developers rely on for real-time voice agents" and that the company is excited to welcome members of the team.
The sequence matters. Inworld already ranks Realtime TTS-2 at the top of the Artificial Analysis text-to-speech leaderboard, with support for more than 200 languages and latency under 100 milliseconds. A strong voice model sells better when it sits inside the product developers use to assemble agents. Ultravox gives it that placement.
The price was withheld, so the premium Inworld paid is unknowable. Ultravox's last disclosed funding was the 2023 seed round, raised when the company operated as Fixie. That round was led by Redpoint Ventures, with Madrona Venture Group, Zetta Venture Partners, SignalFire, Bloomberg Beta and Kearny Jackson also participating. Ultravox raised no disclosed round afterward, and the sale of a company that has not announced a follow-on raise reads as a consolidation outcome.
What Changes for Ultravox Customers?
Customers who use built-in Inworld voices now get Realtime TTS-2 automatically, with no code changes and no additional cost. Existing voice IDs continue to work, and support continues from both Inworld and the original Ultravox builders. New controls cover pace, tone and emotional delivery through inline instructions.
For customers on voices from other providers, the statement was narrower: nothing changes today. Ultravox previously relied on ElevenLabs for most of its internal voices, as of May 2025. Inworld has made no public commitment on pricing, contract terms or how long third-party voices will remain supported. Ultravox's standard terms also allow contract assignment without consent, and material changes take effect 30 days after notice.
Inworld's Position
Inworld was founded in 2021 by Ilya Gelfenbeyn and Kylan Gibbs and began in AI characters for games. It has since moved toward infrastructure for consumer AI applications. Its backers include Lightspeed Venture Partners, Kleiner Perkins, Founders Fund, CRV and Microsoft's M12. Its last widely reported valuation was above $500 million, set in a 2023 round, and total funding is above $117 million, a figure that has been cited as higher than $125 million in more recent coverage.
The Ultravox purchase follows a June 2026 price cut on Inworld's own services, aimed at the cost of running consumer AI at scale. Buying distribution for its voice models fits the same strategy: more developers building on Inworld's stack means more inference volume to amortize.
What Comes Next for Voice Agent Infrastructure?
The next test is whether Ultravox stays neutral on voices and models. Developers building on the platform picked it in part for flexibility, and a vendor owned by a text-to-speech provider has an incentive to favor its own voices. Inworld's decision to upgrade only its own voices at launch points in that direction, though it has not restricted other providers.
Competing voice-agent platforms and speech-model vendors will watch whether Ultravox customers stay after the 30-day change windows in its terms start to apply. A speech-to-speech model trained by the same team that runs the orchestration layer could also narrow latency further. That would raise the bar for rivals that stitch together separate transcription, language and voice components.
Outlook
Inworld paid an undisclosed sum for a developer platform, a team with turn-taking expertise and an open-weight speech-native model. The immediate benefit for customers is a free voice upgrade. The open questions are pricing, third-party voice support and whether Ultravox remains a neutral layer. Those answers will determine whether the deal consolidates Inworld's voice stack or pushes developers toward alternatives.
