Free Speech-to-Text APIs in 2026

A speech-to-text (STT) API turns recorded or live audio into a transcript. Seven verified offers in the catalog let you do that for free, in two shapes: a one-time credit you spend down (Deepgram, AssemblyAI, Speechmatics, Gladia, Scaleway's free minutes) and an allowance that refills (Groq daily, Cartesia monthly). Every number below was checked on the vendor's own pricing or docs page on 2026-10-20; where the vendor did not state a figure, the guide leaves it out.

Compare the free speech-to-text APIs

The hour counts for Deepgram, AssemblyAI and Speechmatics are our arithmetic from the published pay-as-you-go rate; those vendors state the credit and the rate, not the hours. Groq's figures are its free-plan limits of 7,200 audio seconds an hour and 28,800 a day.

Which one to pick

• Most free hours, once: Deepgram's $200 credit, which does not expire and also covers text-to-speech, so one key serves a whole voice app.
• A free allowance that never runs out: Groq's free plan refills every day: up to 8 hours of Whisper audio a day, 20 requests a minute. Good for batch jobs that can wait out a limit.
• Many languages: Gladia (100+ languages, about 312 hours of async audio for EUR 50) or Speechmatics (55+ languages).
• Live captions and voice agents: Deepgram's 150 streaming connections; AssemblyAI's free tier opens only 5 new streams a minute, and Speechmatics allows 2 real-time sessions on Free.
• STT and TTS on one monthly plan: Cartesia's 20,000 credits cover both; for the speech side, see our free text-to-speech API guide.

Steps

1. Pick an offer from the table and create an account. Speechmatics pauses API access when the free credit runs out instead of billing you; add a card only when you want to continue on pay-as-you-go.
2. Create an API key and keep it server-side in an environment variable: anyone holding the key spends your credits.
3. Transcribe one short file first and check the transcript, language detection and timestamps before you send a backlog.
4. Pick pre-recorded (batch) over streaming when you can: on Deepgram and Gladia, and for Speechmatics' Enhanced model, batch audio costs less per hour, so the same credit lasts longer.
5. On Groq, split long recordings and queue requests: the free plan counts audio seconds per hour and per day as well as requests per minute.

Need text-to-speech too?

Deepgram's credit, Speechmatics' credit and Cartesia's monthly credits also cover TTS. The free text-to-speech API guide compares ElevenLabs, Cartesia, Amazon Polly, Deepgram, Hume AI and Sarvam AI, and the free AI voice and audio tools hub lists every audio offer we track.

When to self-host

When credits run out or audio cannot leave your machine, run Whisper (MIT) locally; the catalog links it as the self-hosted stand-in for Deepgram, AssemblyAI, Speechmatics and Gladia. It has no per-hour bill; the tiny and base models need about 1 GB of VRAM and the large model about 10 GB.

Limits and the catch

• Most free STT is a one-time credit. Only Groq and Cartesia refill, and Cartesia's monthly budget is under two hours of audio.
• Expiry varies. Speechmatics' sign-up credits last 12 months and Gladia's one year; Deepgram's do not expire.
• Streaming is the expensive path. Real-time rates are higher than batch at Deepgram and Gladia, and the free tiers cap concurrent streams.
• Free tiers change. Offers are re-checked monthly; the verified date on each offer page is the day we last confirmed it.