Voice, Speech & Realtime AI

Deepgram hits $100M ARR, launches Flux — a conversation-native text-to-speech model

Source: Deepgram · Sep 11, 2026

Deepgram has crossed $100 million in annual recurring revenue and used the milestone to launch Flux, a conversation-native text-to-speech model built specifically for voice agents rather than general narration or audiobook-style synthesis. Flux's pitch is context-aware pacing and emphasis without requiring SSML markup — the model infers where a conversational response should pause, speed up, or stress a word from the text and dialogue context alone. It's free to use through September 12, with standard pricing starting September 13.

This is a meaningful product-line expansion for Deepgram, which built its reputation and this revenue on speech-to-text. Voice agents need both directions — accurate transcription of what a caller says, and natural-sounding, appropriately-paced speech back — and Deepgram now owns both halves of that loop rather than partnering out the TTS half. That's the same infrastructure-to-full-stack pattern we flagged with Deepgram's OfOne acquisition earlier this month: a voice API vendor consolidating the pieces needed to run a complete voice-agent product, not just the recognition layer.

Key Points

• Deepgram has crossed $100M in annual recurring revenue
• Flux is a conversation-native text-to-speech model built for voice agents
• Context-aware pacing and emphasis, no SSML markup required
• Free tier available through September 12; standard pricing begins September 13