Screenshots 2
Deepgram
Summary
Stitching together a separate STT model, a TTS engine, and an LLM layer for a voice agent means three latency budgets, three failure surfaces, and three vendor contracts — Deepgram exists because that stack collapses under real conversational load.
Deepgram provides a Speech-to-Text API, a Text-to-Speech API, and a Voice Agent API that the vendor describes as a single unified pipeline — STT, LLM routing, and TTS through one API call rather than three chained requests. The Flux STT and Flux TTS models are described as conversation-aware, designed to handle turn-taking and interruptions natively. Cloud deployment is the standard path; the vendor mentions self-hosted availability specifically for enterprise custom-model engagements, not as a self-service option. At free-tier volumes the API is accessible immediately, but teams building compliance-sensitive or regulated-industry workflows will hit gating on custom model access and require a sales conversation before deployment.
Bottom line: Deepgram is the right call when you need a low-latency real-time voice agent API and can run in their cloud — it breaks as a path when your compliance team requires on-premise data residency without enterprise contract overhead.
Pricing Plans
Usage-Based- Free Tier
- Pay As You Go with $200 credit; STT up to 50 REST / 150 WSS concurrency; TTS up to 45 REST+WSS; Voice Agent up to 45 WSS; Audio Intelligence up to 10 REST.
Pay As You Go
No minimums. $200 credit then pay-as-you-go. Up to 50 REST / 150 WSS STT concurrency.
- Community & Discord support
- Standard uptime SLA
- All public model endpoints
Growth
Pre-paid credits for the year. Up to 225 WSS STT concurrency and discounts up to 20%.
- Higher concurrency limits
- Community & Discord support
- Standard uptime SLA
Enterprise
Custom volumes, data, deployment, and support needs. Contact sales.
- Custom SLAs and support
- Higher limits
- Compliance options including BAA
View full pricing on deepgram.com →
Pricing may have changed since last verified. Check the official site for current plans.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Unified STT-LLM-TTS pipeline in a single API call, which means teams avoid the compounding latency and error surface of chaining three separate vendor APIs together.
- Conversation-aware Flux models handle interruptions and turn-taking natively, so voice agents don't break when a user cuts off a response — the failure mode that makes generic transcription APIs unusable for live conversational products.
- Real-time and batch modes on the same API, so the same integration handles a live support agent and an overnight compliance transcription job without switching providers.
- Provider-level audio intelligence features — diarization, speaker separation, compliance monitoring — available through the same API surface, so teams avoid bolting on a separate analytics layer for meeting or call-center workloads.
- Freemium entry with pay-as-you-go pricing above the free credit threshold, which means a prototype can go to production without a procurement cycle blocking the first real traffic.
Cons
Sign in to edit- Self-hosted deployment is not a self-service option: the vendor page describes it as part of an enterprise custom-model engagement requiring a sales conversation. Teams with data-residency requirements that cannot route audio through Deepgram's cloud hit this wall before writing a single line of integration code — and at that point they evaluate on-premise-first alternatives like AssemblyAI's self-hosted offering or open-source Whisper deployments.
- Custom model training and enterprise compliance configurations are behind a sales gate, not accessible through the standard API tier. A startup that outgrows the default models and needs domain-specific accuracy — medical terminology, heavy accents, proprietary jargon — faces a contract negotiation rather than a dashboard setting.
- The concurrent streaming connection cap on the Flux TTS free-access period (the vendor states 45 globally, 5 in EU/AU) means load testing a real-time voice product at scale requires moving to paid access before you have production validation of the model's behavior under your specific traffic patterns.
About
- API Available
- Yes
- Self-Hosted
- No
- Last Updated
- 2026-08-14T02:14:25.308Z
Best For
Who it's for
- Developers and startups
- Growing applications
- Enterprise-scale deployments
- Real-time voice solutions
What it does well
- Real-time transcription for applications
- Voice assistants and conversational AI
- Audio analysis for insights and compliance
- Multi-speaker meeting transcription
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Compare Deepgram
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is Deepgram free?
- Deepgram has a permanent free tier alongside paid upgrades. You can keep using a baseline version indefinitely without paying.
- Is Deepgram open source?
- No — Deepgram is a closed-source tool. Source code is not publicly available.
- Does Deepgram have an API?
- Yes. Deepgram exposes a developer API. See the official documentation at https://deepgram.com for details.
Curated lists that include this category
Stitching together a separate STT model, a TTS engine, and an LLM layer for a voice agent means three latency budgets, three failure surfaces, and three vendor contracts. Deepgram supplies a single API call that routes STT, LLM, and TTS together.
Core capabilities
The Flux STT and Flux TTS models handle turn-taking and interruptions natively. Real-time and batch modes run on the same integration. Audio intelligence features support analysis and compliance checks. The vendor states self-hosted options exist only inside enterprise custom-model deals, not as a self-service path.
Pricing and limits
Usage-based pricing starts with a $200 credit on the pay-as-you-go plan. Free-tier concurrency caps are listed as 50 REST or 150 WSS for STT, 45 for TTS and Voice Agent, and 10 for Audio Intelligence.
Who it is for / who should skip it
Developers and startups building real-time voice solutions or multi-speaker transcription fit the listed best-for profile. Teams needing immediate self-hosted deployment or sales-free custom model training should skip it and review on-premise-first alternatives instead.