The most accurate transcription API is also the lowest cost.
95.1% accuracy on the CommonVoice benchmark, above Assembly AI, OpenAI, Azure, Deepgram, and Google, from $0.16 per audio hour. Translation, speaker identification, summaries, sentiment, and captions are included, not upsold.
Benchmark on the CommonVoice 5.1 dataset, replicating Assembly AI's published methodology. Method and results.
Better accuracy, lower cost, against every major API.
Best accuracy, or fastest turnaround.
Transcription API
Built for high-volume batch transcription where accuracy matters. Whisper Large v3 class models with audio enhancement and LLM post-processing.
- No. 1 benchmark accuracy; 8 languages fully supported
- English translation, speaker identification, word-level timing, SRT output
- Summaries, sentiment, classification, and custom LLM prompts
- Custom vocabulary for domain terms
- 5x real-time speed
| Monthly volume | Price / audio hour |
|---|---|
| 0 – 8,333 hrs | $0.20 |
| 8,334 – 16,666 hrs | $0.18 |
| 16,667 – 33,333 hrs | $0.16 |
Transcription Lite
Lower latency, standard accuracy, core languages. For pipelines where speed and cost matter more than the last points of accuracy.
- Standard accuracy; core languages
- English translation, speaker identification, timing, SRT output
- 40x real-time speed
| Monthly volume | Price / audio hour |
|---|---|
| 0 – 8,333 hrs | $0.08 |
| 8,334 – 16,666 hrs | $0.05 |
| 16,667 – 33,333 hrs | $0.03 |
Billed on audio duration to the nearest 0.01 hour (36 seconds). Failed transcriptions are not charged. Volumes above 33,333 hours a month: talk to sales.
Many features. One price.
Speech-to-text
Automatic speech recognition with diarization and word-level time codes, tuned with audio enhancement in pre-processing.
Summaries & sentiment
LLM-generated call summaries, emotion and sentiment detection, and text classification on every transcript.
Translation
Translate 97 languages to English, or full translation between 8 supported languages.
Captions & subtitles
Industry-standard SRT output at scale, for video as well as audio.
Custom vocabulary & prompts
Domain terms, names, and product vocabulary; custom LLM prompts for structured extraction.
Integrations
Zapier and Pabbly Connect bring the API to Zoom, Notion, Dropbox, Google Drive, YouTube, Vimeo, and hundreds more.
What people building on it say.
"A freaking nuclear reactor for adding transcription to your own services. If you need to transcribe-enable your workflows, then Salad is the best I've found."
Dr. Joseph Sweeney · Analyst & Advisor, IRBS"A game-changer for SaaS entrepreneurs and a launchpad for innovation & business growth. Despite its low cost, the transcriptions are remarkably fast and accurate."
Danny McCoy · EntrepreneurSwitching and scaling.
How do I migrate from another provider?
Migration guides cover Deepgram, Assembly AI, Amazon Transcribe, Google, and Azure. Most teams switch in a day; the payback is typically measured in weeks. Start with the docs.
Why is it so much cheaper?
Other APIs rent expensive data-center GPUs from hyperscalers. Salad's transcription runs on the Community Cloud's consumer GPUs, the lowest-cost compute in the market, and tiered pricing passes volume savings through.
Can we self-host it?
Yes. Enterprise customers can license the models for deployment on-prem, in your own cloud, or through Salad Private Cloud. Talk to sales about hosting packages.
How is data handled?
Files are processed in isolated environments on SOC 2 Type I compliant infrastructure with end-to-end encryption, data sanitization, and access controls. Completed jobs and their results are deleted on a fixed schedule.