About this page

Benchmarks with their method attached

Deepgram's benchmarks page opens on a dark textured band with "How Deepgram benchmarks voice AI", a line promising the methodology and testing dates behind every number, and a magenta "Try for free" pill. "STT, TTS, and the test set" switches between three tabs and shows three object cards: Batch and streaming, 20 languages, Accuracy under noise.

A head-to-head table, then a leaderboard

"Deepgram against the field" states the test conditions and dates, a "Word error rate, lower is better" chip, and a table pitting Deepgram against the best competitor for batch, streaming, speech and hard prompts. On the light half, "Measured on real, messy production audio" uses tabs and a bar leaderboard where Deepgram's 13.78 sits in a magenta bar above ten vendors in grey.

Your own test next

Two large stats (13.78% and "6 of 6") lead to a "Run your own benchmarks" card with a green image and "Try now", then six FAQs about benchmark pitfalls.

Sections of this page 1

Similar Features Page screens

All Features Page screens

More from Deepgram

All Deepgram screens