convert.express

Three transcription engines. One choice.

convert.express lets you pick the AI model behind your transcription. Choose OpenAI Whisper for a generally available model with single-pass translation, MAI-Transcribe 1.5 for speed, or Gemini 3.5 Transcribe for the widest language coverage and in-model speaker labels. Same price whichever you pick.

Side-by-side comparison

Any model's transcript can be translated afterward into any supported language using the dashboard's Translate action — the row above only covers Whisper's free, single-pass translation option at upload time.

The three word error rates come from three different benchmarks measured under different conditions — Whisper's from published third-party clean-benchmark and real-world comparisons, MAI-Transcribe 1.5's from the Artificial Analysis Word Error Rate (AA-WER) leaderboard, and Gemini 3.5 Transcribe's from Google's own published figure. They are not measured on the same audio and are not directly comparable to one another. We have not yet measured our own word error rate on a reference transcript, so treat that row as vendor and third-party claims rather than a ranking.

Speed figures for Whisper and MAI-Transcribe 1.5 are vendor and benchmark figures. The Gemini 3.5 Transcribe figure is ours, extrapolated from a 30-minute recording transcribed in 75 seconds on our own pipeline; real times vary with file length, queue load, and whether speaker labeling is on.

OpenAI Whisper

  • Word Error Rate~2.7% clean / ~8% real-world
  • Speed (1 hr audio)~60–90 s
  • Languages supported57
  • Single-pass translation (upload time)English only
  • Entity biasingVocabulary hints (free text)
  • Speaker labelingYes (separate pass)
  • StatusGenerally available
  • Price€0.02 / min

MAI-Transcribe 1.5Preview

  • Word Error Rate2.4% (AA-WER leaderboard)
  • Speed (1 hr audio)< 15 s
  • Languages supported43
  • Single-pass translation (upload time)Not supported
  • Entity biasingUp to 200 phrases
  • Speaker labelingYes (separate pass)
  • StatusPublic preview
  • Price€0.02 / min

Gemini 3.5 TranscribePreview

  • Word Error Rate2.6% (vendor-published)
  • Speed (1 hr audio)~150 s (our measurement)
  • Languages supported60
  • Single-pass translation (upload time)Not supported
  • Entity biasingUp to 1,000 biasing terms
  • Speaker labelingIn-model, up to 8 speakers
  • StatusPublic preview
  • Price€0.02 / min

When to use each model

OpenAI Whisper

  • Production workflows that need a generally available model with no preview caveats
  • Translating foreign-language audio straight to English in one pass, free
  • Maori, Urdu, Welsh — the 3 languages only Whisper covers
  • ~2.7% WER on clean benchmark speech, ~8% on real-world audio

MAI-Transcribe 1.5

Preview
  • Speed-critical workflows — 1 hr of audio in under 15 s
  • 2.4% WER, #3 on the Artificial Analysis (AA-WER) leaderboard
  • Professional use cases: medical, legal, research (entity biasing)
  • Word-level timestamps for precise subtitle generation

Gemini 3.5 Transcribe

Preview
  • The widest language coverage of the three, at 60 languages
  • Interviews and meetings — speakers are labelled in-model, up to 8
  • Long glossaries: up to 1,000 biasing terms for names and jargon
  • Recordings that switch language mid-sentence

About the Preview models

MAI-Transcribe 1.5 is in Microsoft's public preview programme and Gemini 3.5 Transcribe is in Google's. Neither carries a formal SLA. OpenAI Whisper is generally available and stays the default.

If you pick a model explicitly, that is the model your job runs on — we never silently transcribe with a different one, so if your chosen model is unavailable the job fails and tells you why rather than returning a transcript from somewhere else. Jobs left on the default do fall back: Whisper runs first, and MAI-Transcribe 1.5 picks the job up if Whisper cannot take it.

Model selection is in Advanced Options on the upload form — free to try.

Try it now →

See also: language coverage by model · model recommendations by use case