Editorial illustration for AssemblyAI Ships Streaming TTS-1 API for Voice Agents at USD 15 per 1M Characters
AssemblyAI Launches Streaming TTS-1 API at $15/M Characters
AssemblyAI released its first text-to-speech model Tuesday, AssemblyAI-TTS-1, pricing it at $15 per 1 million characters with no annual contract and no concurrency tier to buy. The company built its name on speech-to-text transcription, so the move into voice generation closes a loop for developers who already run AssemblyAI on the listening side of a voice agent. Now they can handle the talking side with one vendor instead of two invoices.
The target use case is live phone calls, where a few hundred milliseconds of lag or a mangled word is the difference between a caller staying on the line and hanging up. AssemblyAI says TTS-1 was trained on millions of hours of audio, much of it real conversational speech rather than scripted narration, and that it works on continuous audio representations instead of the discrete tokens most competing models rely on. The API ships with access to a library of more than 4,000 voices and instant voice cloning, runs only on AssemblyAI's own cloud in the US or EU, and reaches customers over standard HTTP or WebSocket connections. No open weights, no on-prem deployment.
On Cekura’s benchmark of 27 models, TTS-1 scored 81.5% reading accuracy and 4.50 naturalness. Its median streaming latency, timed from the last word sent, was 159 ms. Inworld TTS-2 Flash posted 45 ms and 93.2% accuracy there, at the same $15 price.
Why this matters
For teams building voice agents, the pitch here is consolidation, not novelty. AssemblyAI already has a sizable base of developers using its speech-to-text API, and TTS-1 at $15 per 1M characters with no contract or concurrency tier is clearly aimed at keeping them inside one billing relationship instead of bolting on ElevenLabs or Cartesia. That's a real convenience for founders trying to ship a phone-call agent without stitching together three vendors, but we'd want to see latency and pronunciation benchmarks against the incumbents before assuming this is a straight upgrade, not just a cheaper default.
The continuous-representation claim, versus the discrete tokens most TTS models use, is the one technical detail worth watching: if it actually reduces artifacts on real conversational audio, that's a meaningful differentiator; if it's just framing, the pricing does the talking instead. For researchers, the training emphasis on real conversational data (not just studio audio) is the more interesting signal about where TTS evaluation is heading. Worth testing on an actual live call before switching.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source Box CEO Aaron Levie Says AI Will Drive "Biggest Cybersecurity Upgrade
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives Infini-AI-Lab Releases Hermit, Open-Source Agent Harness Adapter Framework
From Archives GitHub Copilot Projects Adds Task Tracking to Code Assistant
Common Questions Answered
What is the pricing model for AssemblyAI's TTS-1 API?
AssemblyAI-TTS-1 is priced at $15 per 1 million characters with no annual contract required and no concurrency tier to purchase. This straightforward pricing structure eliminates hidden fees and long-term commitments, making it accessible for developers of all scales building voice agents.
How does AssemblyAI-TTS-1 perform compared to competitors like Inworld TTS-2 Flash?
According to Cekura's benchmark of 27 models, TTS-1 achieved 81.5% reading accuracy and 4.50 naturalness with a median streaming latency of 159 ms. While Inworld TTS-2 Flash offers faster latency at 45 ms and higher accuracy at 93.2%, both services are priced identically at $15 per 1 million characters.
Why does AssemblyAI's move into text-to-speech matter for voice agent developers?
By adding TTS-1 to its existing speech-to-text transcription capabilities, AssemblyAI enables developers to handle both the listening and talking sides of voice agents through a single vendor. This consolidation eliminates the need to integrate multiple services like ElevenLabs or Cartesia, simplifying billing and reducing vendor management complexity for teams building phone-call agents.
What is the primary use case for AssemblyAI-TTS-1?
The target use case for TTS-1 is live phone calls, where low latency and natural-sounding speech are critical for real-time voice agent interactions. The streaming API is specifically designed to handle the real-time demands of conversational phone applications.
Further Reading
- AssemblyAI-TTS-1: text-to-speech — built for voice agents - AssemblyAI
- Introducing our Voice Agent API - AssemblyAI
- Best voice agent API in 2026: 6 options compared - AssemblyAI
- The voice AI stack for building agents in 2026 - AssemblyAI
