What's Happening?
Together AI has launched Qwen3 TTS on its serverless inference platform, introducing real-time text-to-speech capabilities for developers. This new integration, announced on November 20, 2026, provides access to an open text-to-speech model featuring
nine preset voices and support for 10 languages, including English, Chinese, Japanese, and Korean. The service is priced at $20 per 1 million characters. A key feature is its low latency, delivering the first audio in as little as 97 milliseconds, which is crucial for real-time voice agent applications. The Qwen3 TTS model, developed by Alibaba Cloud's Qwen team, separates speaker identity from delivery style, allowing developers to control emotional tone through plain-language instructions without altering the chosen voice. This architectural design, combined with HTTP streaming and a WebSocket API, enables the model to handle text arriving in pieces, making it suitable for dynamic conversational AI. Together AI also offers a VoiceFinder tool for developers to preview voices with custom text before implementation.
Why It's Important?
This development is significant for the enterprise AI market, particularly in customer support and experience automation, which is a leading generative AI application among decision-makers. The ability to deploy real-time voice agents with low latency and multilingual support directly addresses the demand for automated service, personalized engagement, and support triage. Enterprises can maintain a consistent brand voice across different languages and markets, reducing operational overhead. Together AI's strategy of offering an open inference stack provides developers with flexibility, allowing them to integrate best-available open models without vendor lock-in, a preference for 51% of enterprise decision-makers who favor a mix of in-house and vendor solutions. This positions Together AI as a comprehensive alternative to proprietary model ecosystems, reinforcing its credibility as a full-stack inference provider. The AI platforms market is projected to reach $1.46 trillion by 2026, indicating substantial growth potential for platforms that can capture developer spend through broad and high-quality offerings.
What's Next?
The immediate future will likely see enterprises integrating Qwen3 TTS into their production deployments for customer experience teams, with adoption scale in Q1 2027 being a key indicator of success. Competitors in the proprietary text-to-speech market are expected to respond to Together AI's 97ms latency benchmark with their own serving optimizations. There is potential for Together AI to expand its language coverage beyond the current 10, targeting markets in Southeast Asia and the Middle East to broaden its global reach. The development of the open-model ecosystem will also be crucial, with expectations for more open TTS models to appear on the Together AI catalog as the voice agent market matures. Furthermore, the preference of 51% of decision-makers for hybrid solutions will be tested as they translate this preference into serverless contracts with Together AI through Q2 2027, indicating the platform's ability to attract and retain enterprise clients.
Beyond the Headlines
The introduction of Qwen3 TTS by Together AI highlights a broader shift towards open-source, flexible AI solutions that empower developers to customize their AI stacks. This move challenges the dominance of proprietary ecosystems by offering a viable, high-performance alternative, potentially democratizing access to advanced AI capabilities. The emphasis on separating speaker identity from delivery style in voice generation has ethical implications, allowing for consistent brand representation while enabling nuanced emotional expression, which could enhance user experience but also raises questions about the authenticity of AI-generated interactions. The asset-light approach to model acquisition, where Together AI integrates models developed by others, signifies a trend towards platform strategies focused on inference quality and breadth rather than exclusive model ownership. This collaborative model could accelerate innovation across the AI landscape, fostering a more diverse and competitive market for generative AI services.













