SpeechWeave vs Deepgram: lower-cost transcription for recorded audio
Built for pre-recorded pipelines: lower batch rates, drop-in OpenAI SDK compatibility, and audio deleted after every job.
Rates and limits verified October 2026.
| SpeechWeave | Deepgram | SpeechWeave advantage | |
|---|---|---|---|
| Price per hour | Core standard: $0.09/hr Max standard: $0.24/hr Core deferred: $0.045/hr Max deferred: $0.12/hr | Nova-3 monolingual: $0.258/hr Nova-3 multilingual: about $0.312/hr | Up to 65% lower than Nova-3 monolingual, and deferred mode cuts it further |
| Max file size | 512 MB per job, higher limits available on request | Up to about 2 GB (pre-recorded API) | |
| Integration | SpeechWeave SDKs (Deepgram-style helper), OpenAI SDK, or REST | Deepgram SDKs and API | Keep your Deepgram call and response shape |
| MCP server for Cursor / Claude | Official @speechweave/mcp | Not a first-party MCP focus | AI-assisted workflows out of the box |
Keep your Deepgram call shape
The SpeechWeave SDKs include a Deepgram-style prerecorded helper. The call and the nested response keep the same shape, so your existing parsing code stays as it is.
JavaScript
import { SpeechWeave } from "@speechweave/node";
const sw = new SpeechWeave({ api_key: process.env.SPEECHWEAVE_API_KEY });
const dg = await sw.listen.prerecorded.transcribeUrl("https://example.com/audio.mp3", {
model: "core",
});
console.log(dg.results?.channels?.[0]?.alternatives?.[0]?.transcript);Python
import os
from speechweave import SpeechWeave
sw = SpeechWeave(api_key=os.environ["SPEECHWEAVE_API_KEY"])
dg = sw.listen.prerecorded.transcribe_url(
"https://example.com/audio.mp3",
model="core",
)
print(dg["results"]["channels"][0]["alternatives"][0]["transcript"])Prefer the OpenAI SDK or plain REST? SpeechWeave also exposes an OpenAI-compatible transcription endpoint.
Why teams with recorded audio choose SpeechWeave
- Lower rates: Core $0.09/hr, deferred $0.045/hr, against $0.258/hr for Nova-3 monolingual.
- Familiar call shape: The SpeechWeave SDKs include a Deepgram-style prerecorded helper that returns the same response structure. You can also use the OpenAI SDK or plain REST, or connect Cursor and Claude through @speechweave/mcp.
- Work that can wait: Deferred service mode runs background jobs at half the standard rate.
- Predictable spend: A prepaid wallet with spending caps and auto-recharge, so a spike in uploads never becomes a surprise bill.
- Audio privacy: Audio is deleted after every job and never used to train models.
SpeechWeave is purpose-built for recorded audio. If your product handles recorded calls, voicemails, webinars, or archives, SpeechWeave gives you better margins.