Deepgram
← Back to All AI Tools1. Should you try Deepgram? Overview
| What it is | Deepgram is a speech-to-text API built for developers who need to transcribe a lot of audio accurately and cheaply — calls, meetings, podcasts, and video. |
| Country / HQ | San Francisco, USA |
| Best for | engineering teams transcribing at scale |
| Check before buying | Sign up and grab the $200 free credit — enough to transcribe hundreds of hours at evaluation rates. |
| Pricing signal | Free tier + paid from $200/mo |
2. Features What it does
| Feature | What it does |
|---|---|
| Speech | to-text API with streaming and pre-recorded modes. |
| Multiple | models — Nova-3 (flagship), Nova-2, and Flux for lower-latency needs. |
| Per | second billing — you pay for actual audio length, not rounding. |
| Self | hosted option for keeping audio on your own infrastructure. |
| Diarization, | summarization, and language detection add-ons. |
3. Pricing Pricing (2026-10, USD)
| Plan | Price / mo | Key limits & who it fits |
|---|---|---|
| Nova-3 (streaming, mono) | $0.29/hr | English / single-language streaming. |
| Nova-3 (multilingual) | $0.35/hr | Many languages in one stream. |
| Flux | $0.39-$0.47/hr | Lower-latency variant. |
| Growth plan | $4,000/yr min | Discounted rates, annual commitment. |
| Free credit | $200 | New accounts; roughly 700 hours at Nova-3. |
Note: prices are per audio hour and billed per second, so a 90-second clip costs a quarter of an hour’s rate. The free $200 credit is generous for evaluation.
4. Integration Access & integrations
| Access type | Details |
|---|---|
| REST + WebSocket API + SDK + self-host | Python/Node SDKs, webhooks, Zapier — Any backend, on-prem |
5. Reviews & Sources External reviews
| Source | Rating | What it says |
|---|---|---|
| Pros | — | Among the cheapest accurate APIs once you are past the free credit. |
| Pros | — | Per-second billing and a real free credit make cost predictable. |
| Pros | — | Self-host and on-prem option is rare among STT vendors. |
| Cons | — | No turnkey consumer app — you need to build on the API. |
| Cons | — | Accuracy on heavy accents and niche jargon still trails human transcription. |
| Cons | — | Growth plan’s $4K/yr floor is a jump from pure pay-as-you-go for light users. |
6. Best for / Not for Fit
| ✔ Suitable | ✘ Not suitable |
|---|---|
| • Developers transcribing calls, meetings, or media at volume. • Teams with data-residency needs that want self-hosting. |
• Non-technical users wanting a ready-made transcription app. • Buyers needing speaker-labeled, fully-punctuated output with zero cleanup. |
7. Competitors How it compares
| Competitor | How it differs | Best for |
|---|---|---|
| Deepgram | API + app: API only; Per-second bill: Yes; Free credit: $200 | Cheap scale |
| AssemblyAI | API + app: API only; Per-second bill: Tiered; Free credit: Free trial | Feature-rich |
| Whisper (OpenAI) | API + app: Self-host model; Per-second bill: Free (own GPU); Free credit: $0 | Privacy/own |
| Speechmatics | API + app: API + self-host; Per-second bill: Yes; Free credit: Trial | Accuracy |
8. Summary Verdict
| Takeaway | Deepgram is the cost-efficient default for engineering teams that need to transcribe a lot of audio through an API — per-second billing, a genuine $200 free credit, and a self-host option that most STT vendors do not offer. It is not for non-developers with no app to use. |
| Who should pick it | Developers transcribing calls, meetings, or media at volume. |
| Who should skip it | Non-technical users wanting a ready-made transcription app. |