Ranker Models Playground

Speech, decoded.

One API to every speech-to-text model worth using. Compare side by side, route to the best, pay by the second.

How it works · 3 steps

From audio to transcript in seconds.

Upload audio

Drop any audio file — MP3, WAV, FLAC, and 10 more formats.

Drop audio file

Pick a model — or let us

Choose from 30+ models or use our smart router to auto-select by price, speed, or accuracy.

Whisper Large v3

Nova 3

Chirp 2

Get a structured transcript

Receive word-level timestamps, speaker labels, and confidence scores in a unified format.

{
  "start": 0.0,
  "end": 2.4,
  "text": "Welcome to...",
  "speaker": "A",
  "confidence": 0.97
}

Features

Everything you'd build anyway.

The stuff that's table-stakes the day you ship a voice product.

One API for every model. Diarization. 100+ languages. Realtime streams. Bring-your-own keys. Real benchmarks. Six things, all built in.

01 / 06

One API, every model

30+ models from 15 providers through a single unified endpoint. One integration, all the options.

provider : "deepgram" →

"assemblyai"

response_shape :{ transcript, words[], speakers[], confidence }

endpoint : POST /v1/transcribe // unchanged

01 / 06

One API, every model

30+ models from 15 providers through a single unified endpoint. One integration, all the options.

provider : "deepgram" →

"assemblyai"

response_shape :{ transcript, words[], speakers[], confidence }

endpoint : POST /v1/transcribe // unchanged

02 / 06

Realtime + batch

WebSocket streaming from mic or file upload. Async batch for pre-recorded audio. Same API.

01:07

~120ms

02 / 06

Realtime + batch

WebSocket streaming from mic or file upload. Async batch for pre-recorded audio. Same API.

01:07

~120ms

03 / 06

Speaker diarization

Identify who said what. Available on supported models with a single flag.

0:02

Quick update — the streaming refactor merged last night.

0:05

Nice. Any change to the latency numbers?

0:09

Down about 15%. I'll post the graph.

03 / 06

Speaker diarization

Identify who said what. Available on supported models with a single flag.

0:02

Quick update — the streaming refactor merged last night.

0:05

Nice. Any change to the latency numbers?

0:09

Down about 15%. I'll post the graph.

04 / 06

100+ languages

From English to Yoruba, with auto-detect on verified models. Code-switching for mixed-language audio.

EN

ES

DE

FR

JA

ZH

AR

HI

PT

RU

KO

IT

TR

PL

NL

SV

TH

VI

+ 86 more

04 / 06

100+ languages

From English to Yoruba, with auto-detect on verified models. Code-switching for mixed-language audio.

EN

ES

DE

FR

JA

ZH

AR

HI

PT

RU

KO

IT

TR

PL

NL

SV

TH

VI

+ 86 more

05 / 06

Bring your own key

Use your existing provider API keys. 100 free minutes per month, 5% platform fee after.

Deepgram

················3f9a

pass-through

AssemblyAI

················a017

idle

05 / 06

Bring your own key

Use your existing provider API keys. 100 free minutes per month, 5% platform fee after.

Deepgram

················3f9a

pass-through

AssemblyAI

················a017

idle

06 / 06

Real benchmarks

WER scores on golden-set audio across 8 categories. Event-driven, not vibes.

WER

06 / 06

Real benchmarks

WER scores on golden-set audio across 8 categories. Event-driven, not vibes.

WER

One-call integration

One call. We post the transcript back.

The SDKs handle the upload and the wait, in TypeScript or Python. Or call the HTTP API directly.

// npm install @opentranscription/sdk
import { readFile } from 'node:fs/promises';
import { OpenTranscription } from '@opentranscription/sdk';

const ot = new OpenTranscription({ apiKey: process.env.OT_API_KEY! });

// Uploads the audio and opens the job. The signed-URL round trip is inside.
const job = await ot.transcribe({
  file: await readFile('interview.mp3'),
  fileName: 'interview.mp3',
  model: 'auto/best',
  language: 'en',
  customWords: ['Kubernetes', 'Grafana', 'Sanjay Bhattacharya'],
  webhookUrl: 'https://your-app.com/webhooks/ot',
});

// We POST a signed event to that URL when the job finishes: no polling required.

↳ auto/best routes to the most accurate model by default.

API reference

From the blog

Deep dives on speech-to-text accuracy, latency and pricing.

All posts

Amazon Transcribe Medical: what AWS actually ships, and what it won't tell you

What Amazon Transcribe Medical offers in 2026: features, specs, pricing vs Google and Nuance, HIPAA posture, research clues, and where it falls short.

Chirp 3: inside Google Cloud's 2025 speech stack, from HD voices to transcription

What Google Cloud Chirp 3 actually is: release timeline, WER and Elo benchmarks, pricing, specs, known issues, and how it compares to Azure and ElevenLabs.

Deepgram Base in 2026: what the legacy model still does well

Where Deepgram Base fits in 2026: API behavior, variants, latency, specs, limitations, and when to pick Nova-3 or Flux instead.

Speech, decoded.

One API to every speech-to-text model worth using. Compare side by side, route to the best, pay by the second.

Compare models on your audio →Start building

$0.0007/min

lowest price

43+

models

105

languages

~772ms

lowest latency

Every model worth using

Deepgram

Google Cloud

Speechmatics

OpenAI

Cartesia

Alibaba (Qwen)

Rev AI

The Ranker

Objective benchmarks. No favorites.

We run automated WER benchmarks against golden-set audio across 8 categories — general, medical, legal, technical, conversational, noisy, finance, and code-switching. Benchmarks trigger on every model or pricing change.

Live ranker

#	Model	Score	WER	Latency	Cost/min
01	Ink-Whisper Cartesia	84.6	17.2%	773ms	$0.0022
02	Whisper Large V3 Turbo Groq	81.3	14.3%	3.4s	$0.0007
03	GPT-4o Mini Transcribe OpenAI	81.1	13.6%	2.0s	$0.003
04	Voxtral Mini Transcribe Mistral (Voxtral)	80.7	12.7%	2.3s	$0.003
05	Whisper Large V3 Groq	79.9	14.4%	3.1s	$0.0019

Ink-Whisper

Cartesia

84.6

WER: 17.2%

Latency: 773ms

Cost/min: $0.0022

Whisper Large V3 Turbo

Groq

81.3

WER: 14.3%

Latency: 3.4s

Cost/min: $0.0007

GPT-4o Mini Transcribe

OpenAI

81.1

WER: 13.6%

Latency: 2.0s

Cost/min: $0.003

Voxtral Mini Transcribe

Mistral (Voxtral)

80.7

WER: 12.7%

Latency: 2.3s

Cost/min: $0.003

Whisper Large V3

Groq

79.9

WER: 14.4%

Latency: 3.1s

Cost/min: $0.0019

+38 more models

View full ranker

+38 more models

View full ranker

Ranked by a composite of accuracy, cost, and speed. See methodology

How it works · 3 steps

From audio to transcript in seconds.

Upload audio

Drop any audio file — MP3, WAV, FLAC, and 10 more formats.

Drop audio file

Pick a model — or let us

Choose from 30+ models or use our smart router to auto-select by price, speed, or accuracy.

Whisper Large v3

Nova 3

Chirp 2

Get a structured transcript

Receive word-level timestamps, speaker labels, and confidence scores in a unified format.

{
  "start": 0.0,
  "end": 2.4,
  "text": "Welcome to...",
  "speaker": "A",
  "confidence": 0.97
}

Features

Everything you'd build anyway.

The stuff that's table-stakes the day you ship a voice product.

One API for every model. Diarization. 100+ languages. Realtime streams. Bring-your-own keys. Real benchmarks. Six things, all built in.

01 / 06

One API, every model

30+ models from 15 providers through a single unified endpoint. One integration, all the options.

provider : "deepgram" →

"assemblyai"

response_shape :{ transcript, words[], speakers[], confidence }

endpoint : POST /v1/transcribe // unchanged

01 / 06

One API, every model

30+ models from 15 providers through a single unified endpoint. One integration, all the options.

provider : "deepgram" →

"assemblyai"

response_shape :{ transcript, words[], speakers[], confidence }

endpoint : POST /v1/transcribe // unchanged

02 / 06

Realtime + batch

WebSocket streaming from mic or file upload. Async batch for pre-recorded audio. Same API.

01:07

~120ms

02 / 06

Realtime + batch

WebSocket streaming from mic or file upload. Async batch for pre-recorded audio. Same API.

01:07

~120ms

03 / 06

Speaker diarization

Identify who said what. Available on supported models with a single flag.

0:02

Quick update — the streaming refactor merged last night.

0:05

Nice. Any change to the latency numbers?

0:09

Down about 15%. I'll post the graph.

03 / 06

Speaker diarization

Identify who said what. Available on supported models with a single flag.

0:02

Quick update — the streaming refactor merged last night.

0:05

Nice. Any change to the latency numbers?

0:09

Down about 15%. I'll post the graph.

04 / 06

100+ languages

From English to Yoruba, with auto-detect on verified models. Code-switching for mixed-language audio.

EN

ES

DE

FR

JA

ZH

AR

HI

PT

RU

KO

IT

TR

PL

NL

SV

TH

VI

+ 86 more

04 / 06

100+ languages

From English to Yoruba, with auto-detect on verified models. Code-switching for mixed-language audio.

EN

ES

DE

FR

JA

ZH

AR

HI

PT

RU

KO

IT

TR

PL

NL

SV

TH

VI

+ 86 more

05 / 06

Bring your own key

Use your existing provider API keys. 100 free minutes per month, 5% platform fee after.

Deepgram

················3f9a

pass-through

AssemblyAI

················a017

idle

05 / 06

Bring your own key

Use your existing provider API keys. 100 free minutes per month, 5% platform fee after.

Deepgram

················3f9a

pass-through

AssemblyAI

················a017

idle

06 / 06

Real benchmarks

WER scores on golden-set audio across 8 categories. Event-driven, not vibes.

WER

06 / 06

Real benchmarks

WER scores on golden-set audio across 8 categories. Event-driven, not vibes.

WER

One-call integration

One call. We post the transcript back.

The SDKs handle the upload and the wait, in TypeScript or Python. Or call the HTTP API directly.

// npm install @opentranscription/sdk
import { readFile } from 'node:fs/promises';
import { OpenTranscription } from '@opentranscription/sdk';

const ot = new OpenTranscription({ apiKey: process.env.OT_API_KEY! });

// Uploads the audio and opens the job. The signed-URL round trip is inside.
const job = await ot.transcribe({
  file: await readFile('interview.mp3'),
  fileName: 'interview.mp3',
  model: 'auto/best',
  language: 'en',
  customWords: ['Kubernetes', 'Grafana', 'Sanjay Bhattacharya'],
  webhookUrl: 'https://your-app.com/webhooks/ot',
});

// We POST a signed event to that URL when the job finishes: no polling required.

↳ auto/best routes to the most accurate model by default.

API reference

Pricing · transparent

Pay per second. No subscriptions.

One rate card across every provider. The price you see is the price you pay. BYOK users pay their provider directly.

Start free

Every new account gets 10 free transcriptions, up to 2 hours of audio — no credit card required.

Compare models on your audio →

Pay as you go

From

$0.0007

/ minute

Per-second billing. Top up credits anytime and spend them across every provider.

Every model included
No monthly minimum
Pay only for audio processed

Get an API key

BYOK

routing fee

Already have provider accounts? Bring your keys, pay your provider directly — we add a 5% routing fee on top.

Your keys, your provider billing
Same playground + ranker
First 100 min/mo free, then 5%

Set up BYOK

Enterprise

Custom

For teams with large volumes, custom deployment, or support requirements.

Volume pricing
Custom deployment options
Invoicing + custom terms

Talk to sales

No subscriptions, no minimums.

From the blog

Deep dives on speech-to-text accuracy, latency and pricing.

All posts

Amazon Transcribe Medical: what AWS actually ships, and what it won't tell you

What Amazon Transcribe Medical offers in 2026: features, specs, pricing vs Google and Nuance, HIPAA posture, research clues, and where it falls short.

Chirp 3: inside Google Cloud's 2025 speech stack, from HD voices to transcription

What Google Cloud Chirp 3 actually is: release timeline, WER and Elo benchmarks, pricing, specs, known issues, and how it compares to Azure and ElevenLabs.

Deepgram Base in 2026: what the legacy model still does well

Where Deepgram Base fits in 2026: API behavior, variants, latency, specs, limitations, and when to pick Nova-3 or Flux instead.

Ready to transcribe?

One API key. No contracts. No minimum spend.

Compare models on your audio →Get an API key