Deepgram
Nova-3 speech-to-text API — real-time and batch transcription at low latency
Deepgram is a speech-to-text API provider specialising in low-latency audio transcription. Its Nova-3 model delivers real-time streaming transcription with high accuracy across accents and domains. Nova-3 Medical is tuned for clinical vocabulary. Deepgram also hosts Whisper Large for teams that need an open-weights baseline. Pricing is per audio hour rather than per token.
Purpose-built speech-to-text API with Nova-3 at ~300 ms latency and a medical-domain variant — priced per audio hour rather than per token.
- Nova-3 delivers industry-leading accuracy with ~300 ms P50 latency for real-time streaming
- Nova-3 Medical tuned for clinical and healthcare vocabulary
- Per-audio-hour pricing is predictable for high-volume transcription workloads
- Supports 30+ languages and automatic language detection
- Whisper Large available for open-weights baseline comparisons
- Speech-to-text only — no LLM chat, image, or embedding models
- No per-token pricing; cost structure differs from general-purpose LLM APIs
- Smaller ecosystem than general-purpose providers
Deepgram Models
| Model | Modality | Tier | Price | Cached /1M | TTFT | Throughput | Context | Intel. | Caps | |
|---|---|---|---|---|---|---|---|---|---|---|
nova-3 Nova | Speech | advanced | $0.22 per hour EST Aug 2026 | — | 300ms | 0 t/s | — | 0 | ||
nova-3-medical Nova | Speech | advanced | $0.43 per hour EST Aug 2026 | — | 350ms | 0 t/s | — | 0 | ||
whisper-large Whisper | Speech | efficient | $0.22 per hour EST Aug 2026 | — | 1.2s | 0 t/s | — | 0 |
Deepgram — Frequently Asked Questions (FAQs)
Similar Providers
Deepgram Reviews
All LLM Providers
Full comparison tableDeepgram LLM pricing overview
Deepgram is a proprietary AI lab that develops and hosts its own large language models. Deepgram is a speech-to-text API provider specialising in low-latency audio transcription. Its Nova-3 model delivers real-time streaming transcription with high accuracy across accents and domains. Nova-3 Medical is tuned for clinical vocabulary. Deepgram also hosts Whisper Large for teams that need an open-weights baseline. Pricing is per audio hour rather than per token. Common use cases include Transcription, Real-time captions, Voice agents, Medical documentation. Headquartered in San Francisco, CA, founded 2015. All models are accessible via a REST API compatible with standard OpenAI-style request formats, enabling drop-in integration with most LLM frameworks and orchestration tools.
Deepgram vs other LLM providers
Deepgram competes with OpenAI, Anthropic, Google, Groq, Together AI, Mistral, Cohere, and other inference API providers across dimensions of price, throughput, latency, context window, and model intelligence. Purpose-built speech-to-text API with Nova-3 at ~300 ms latency and a medical-domain variant — priced per audio hour rather than per token. Use the LLM provider comparison tool to see Deepgram token pricing, latency, and throughput side-by-side with any other provider. The full LLM pricing table shows all providers ranked by input token cost, output token cost, and throughput in a single sortable view.
Why choose Deepgram?
Deepgram stands out because: Purpose-built speech-to-text API with Nova-3 at ~300 ms latency and a medical-domain variant — priced per audio hour rather than per token.. The API currently offers 3 models, spanning 2 advanced, 1 efficient tiers. For teams running high-volume inference workloads, prompt caching and batch API endpoints can reduce effective input token costs by 50–90% — check the context window cost guide for a full breakdown of caching economics.
Understanding Deepgram token pricing
Deepgram charges separately for input (prompt) and output (completion) tokens, priced per 1M tokens in USD. Output tokens are typically 3–5× more expensive than input tokens due to the compute cost of autoregressive generation. Prices shown are sourced from Deepgram's public pricing page and updated every 15 minutes. Need help estimating your spend? Read the LLM API cost calculator guide — it covers tokens, context windows, prompt caching, and batch discounts with worked examples for RAG, chat history, and document processing workloads.
Deepgram context window and model capabilities
Context window size directly affects both capability and cost — every token in the context window is charged as an input token. Deepgram's largest context window is 0 tokens. For RAG and document processing workloads, longer context windows enable richer retrieval but increase per-call costs proportionally. Prompt caching — where supported — stores the KV state of repeated prefixes and charges 75–90% less for cache hits, making it the most impactful cost optimization for applications with consistent system prompts or retrieved documents. See the LLM context window cost guide for a full analysis of how context length affects your API bill.
Self-hosted vs managed inference: when Deepgram makes sense
Managed inference APIs like Deepgram eliminate infrastructure overhead — no GPU provisioning, driver management, or model serving stack to maintain. The trade-off is cost at scale: a single H100 at ~$2.50/hr can serve ~500K tokens/min of Llama 3.3 70B, which at Deepgram API rates would cost significantly more per token. The break-even point depends on your request volume, latency requirements, and engineering capacity. For teams processing fewer than ~10M tokens/day, managed APIs are almost always cheaper when total cost of ownership is considered. Above that threshold, self-hosted inference on rented GPU compute typically wins on unit economics. Read the cheapest GPU cloud guide for a full break-even analysis. Historical Deepgram token price data is available in the LLM price history charts.
Compare Deepgram With Another LLM Provider
Compare Deepgram vs other providers
Side-by-side token pricing, latency, throughput, and context window. View all LLM provider comparisons →