BroadRouter

Speech-to-text API price index

Updated

In checks from 2026-11-14 to 2026-11-15, listed batch speech-to-text prices ran from USD 0.012 to USD 0.96 per audio hour, with a median of USD 0.20. The cheapest Arabic-capable option was Soniox (async file and real-time) at USD 0.10. Streaming cost a median 1.12× the batch price. Every value is sourced and downloadable.

Download CSV Download JSON Croissant metadata Compilation CC BY 4.0 · prices are the providers' published facts · checked 2026-11-14 to 2026-11-15

BroadRouter is BroadNet's own site. BroadNet's two transcription models are included and ranked with the same rules as everyone else; where another provider is cheaper, the tables say so.

Key findings

  1. The cheapest listed batch price is USD 0.012 per audio hour (DeepInfra openai/whisper-large-v3-turbo); the dearest is USD 0.96 (Google Cloud Speech-to-Text V2, standard models (incl. Chirp)), across 21 current USD-priced products from 14 providers.
  2. The median batch price is USD 0.20 per audio hour.
  3. Of the 14 products whose provider names Arabic support, the cheapest batch price is USD 0.10 per hour (Soniox (async file and real-time)); the median is USD 0.255. For 6 products we could not confirm Arabic on the provider's own pages.
  4. Where a provider lists both, streaming costs a median 1.12× the batch price (12 products). 3 charge the same for both; the largest gap is Microsoft Azure AI Speech, standard (East US) at 5.56×.
  5. BroadNet (this site's owner) ranks 6 of 21 on batch price with BroadNet Transcribe at USD 0.12 per hour, and 13 of 21 with BroadNet Transcribe Arabic at USD 0.252. 5 products list a lower price: DeepInfra openai/whisper-large-v3-turbo, DeepInfra openai/whisper-large-v3, Groq whisper-large-v3-turbo, Soniox (async file and real-time) and Groq whisper-large-v3.
  6. Among Arabic-capable products, BroadNet Transcribe ranks 2 of 14; Soniox (async file and real-time) is cheaper. BroadNet Transcribe Arabic ranks 7 of 14. BroadNet lists no streaming price.
  7. 6 products include speaker diarization in the price, 3 charge extra, and 12 do not say. 10 state per-second billing; Rev.ai (15 seconds minimum) and Groq (10 seconds minimum) bill a minimum per request.

How much does speech-to-text cost per hour?

Current USD-priced products, cheapest batch price first. Batch is the standard pre-recorded (file) price at pay-as-you-go list rates, first volume tier.

#Provider and modelBatch per hourStreaming per hourArabicDiarizationSource
1DeepInfra openai/whisper-large-v3-turboUSD 0.012None listedNot statedNot stateddeepinfra.com
2DeepInfra openai/whisper-large-v3USD 0.027None listedNot statedNot stateddeepinfra.com
3Groq whisper-large-v3-turboUSD 0.04None listedNot statedNot statedconsole.groq.com
4Soniox (async file and real-time)USD 0.10USD 0.12YesIncludedsoniox.com
5Groq whisper-large-v3USD 0.111None listedNot statedNot statedconsole.groq.com
6BroadNet TranscribeUSD 0.12None listedYesNot statedinference.broadnet.ai
7AssemblyAI Universal-2USD 0.15USD 0.15YesExtraassemblyai.com
8Google Cloud Speech-to-Text V2, dynamic batchUSD 0.18None listedYesNot statedcloud.google.com
9Microsoft Azure AI Speech, standard (East US)USD 0.18USD 1.00YesExtraprices.azure.com
10ElevenLabs Scribe v3 / v2USD 0.20USD 0.22YesNot statedelevenlabs.io
11Rev.ai Reverb Transcription (English)USD 0.20None listedNoNot statedrev.ai
12AssemblyAI Universal-3.5 ProUSD 0.21USD 0.45Not statedExtraassemblyai.com
13BroadNet Transcribe ArabicUSD 0.252None listedYesNot statedinference.broadnet.ai
14Deepgram Nova-3, monolingualUSD 0.258USD 0.288 (promo)YesIncludeddeepgram.com
15OpenAI gpt-transcribeUSD 0.27USD 1.02Not statedNot stateddevelopers.openai.com
16Rev.ai Reverb Foreign Language TranscriptionUSD 0.30None listedYesNot statedrev.ai
17Deepgram Nova-3, multilingualUSD 0.312USD 0.348 (promo)YesIncludeddeepgram.com
18Amazon TranscribeUSD 0.36USD 0.60YesIncludedaws.amazon.com
19Speechmatics Standard accuracy (Pro, pay as you go)USD 0.45USD 0.45YesIncludedspeechmatics.com
20Speechmatics Enhanced accuracy (Pro, pay as you go)USD 0.75USD 0.80YesIncludedspeechmatics.com
21Google Cloud Speech-to-Text V2, standard models (incl. Chirp)USD 0.96USD 0.96YesNot statedcloud.google.com

Ties share a rank in the findings. 3 deprecated models and 2 euro-priced products are listed in the full table below and in the downloads, but not ranked.

Which speech-to-text APIs support Arabic, and at what price?

"Yes" only where the provider's own language page names Arabic for that model. Dialects are listed as the provider lists them.

Provider and modelBatch per hourArabic locales or dialects listed
Soniox (async file and real-time)USD 0.10Arabic (ar) is in the supported-languages table; no dialect note.
BroadNet TranscribeUSD 0.1214 languages, including Arabic (ar).
AssemblyAI Universal-2USD 0.15Universal-2 supports 99 languages; Arabic is in its "Good accuracy (>10% to ≤25% WER)" group.
Google Cloud Speech-to-Text V2, dynamic batchUSD 0.1818 locales: ar-AE, ar-BH, ar-DZ, ar-EG, ar-IL, ar-IQ, ar-JO, ar-KW, ar-LB, ar-MA, ar-MR, ar-OM, ar-PS, ar-QA, ar-SA, ar-SY, ar-TN, ar-YE
Microsoft Azure AI Speech, standard (East US)USD 0.1818 locales: ar-AE, ar-BH, ar-DZ, ar-EG, ar-IL, ar-IQ, ar-JO, ar-KW, ar-LB, ar-LY, ar-MA, ar-OM, ar-PS, ar-QA, ar-SA, ar-SY, ar-TN, ar-YE
ElevenLabs Scribe v3 / v2USD 0.20Arabic (ara) is listed; ElevenLabs groups it under "Good (>10% to ≤20% WER)".
BroadNet Transcribe ArabicUSD 0.252Arabic only. BroadNet describes it as covering Arabic "including dialects"; its docs do not list dialects.
Deepgram Nova-3, monolingualUSD 0.25817 locales: ar, ar-AE, ar-SA, ar-QA, ar-KW, ar-SY, ar-LB, ar-PS, ar-JO, ar-EG, ar-SD, ar-TD, ar-MA, ar-DZ, ar-TN, ar-IQ, ar-IR
Rev.ai Reverb Foreign Language TranscriptionUSD 0.30Arabic (ar) listed for the Async API.
Deepgram Nova-3, multilingualUSD 0.312Arabic is in the nova-3 multilingual language list; dialects not itemised.
Amazon TranscribeUSD 0.36ar-AE (Gulf), ar-SA (Modern Standard)
Speechmatics Standard accuracy (Pro, pay as you go)USD 0.45Modern Standard Arabic, Gulf, Egyptian, Levantine
Speechmatics Enhanced accuracy (Pro, pay as you go)USD 0.75Modern Standard Arabic, Gulf, Egyptian, Levantine
Google Cloud Speech-to-Text V2, standard models (incl. Chirp)USD 0.9618 locales: ar-AE, ar-BH, ar-DZ, ar-EG, ar-IL, ar-IQ, ar-JO, ar-KW, ar-LB, ar-MA, ar-MR, ar-OM, ar-PS, ar-QA, ar-SA, ar-SY, ar-TN, ar-YE
Gladia Solaria-1 (self-serve)EUR 0.27Languages table: Arabic (ar) supported on Solaria-1, not on Solaria-3.

Not confirmed on the provider's pages: OpenAI gpt-transcribe, OpenAI gpt-4o-transcribe, OpenAI gpt-4o-mini-transcribe, OpenAI whisper-1, AssemblyAI Universal-3.5 Pro, Groq whisper-large-v3, Groq whisper-large-v3-turbo, DeepInfra openai/whisper-large-v3 and DeepInfra openai/whisper-large-v3-turbo. Several are Whisper-based and may work in Arabic; test before you rely on them. For accuracy rather than price, see the Arabic speech-to-text buyer's guide.

How much more does real-time transcription cost?

Streaming price ÷ batch price, for products that list both. Above 1.00× means real time costs more.

Provider and modelStreaming productBatchStreamingPremium
Microsoft Azure AI Speech, standard (East US)Real-time transcriptionUSD 0.18USD 1.005.56×
OpenAI gpt-transcribegpt-live-transcribeUSD 0.27USD 1.023.78×
Gladia Solaria-1 (self-serve)Real-time (self-serve)EUR 0.27EUR 0.682.52×
AssemblyAI Universal-3.5 ProUniversal-3.6 Pro RealtimeUSD 0.21USD 0.452.14×
Amazon TranscribeAmazon Transcribe streamingUSD 0.36USD 0.601.67×
Soniox (async file and real-time)Real-timeUSD 0.10USD 0.121.20×
Deepgram Nova-3, monolingualNova-3 streaming, monolingualUSD 0.258USD 0.288 (promo; regular USD 0.462)1.12×
Deepgram Nova-3, multilingualNova-3 streaming, multilingualUSD 0.312USD 0.348 (promo; regular USD 0.552)1.12×
ElevenLabs Scribe v3 / v2Scribe v3 Live / Scribe v2 RealtimeUSD 0.20USD 0.221.10×
Speechmatics Enhanced accuracy (Pro, pay as you go)Real-time, EnhancedUSD 0.75USD 0.801.07×
Google Cloud Speech-to-Text V2, standard models (incl. Chirp)Same Recognition SKU (no separate streaming price listed)USD 0.96USD 0.961.00×
AssemblyAI Universal-2Universal-Streaming MultilingualUSD 0.15USD 0.151.00×
Speechmatics Standard accuracy (Pro, pay as you go)Real-time, StandardUSD 0.45USD 0.451.00×

Google Cloud lists one V2 recognition price with no separate streaming rate, so its premium is 1.00×. AssemblyAI bills streaming on session time, not audio sent.

Billing increments, minimums and free tiers

Rounding can matter more than the list price for short clips: a 5-second voice note billed with a 15-second minimum costs three times its length.

Provider and modelIncrementMinimumFree tierNotes
OpenAI gpt-transcribeNot statedNot statedNot stated · No free tier stated for transcription.Not stated for gpt-transcribe. A separate diarization model (gpt-4o-transcribe-diarize) is deprecated. OpenAI labels its per-minute figures an estimated cost. Its speech-to-text guide calls gpt-transcribe the recommended model for recorded speech, and its deprecations page names it (or gpt-live-transcribe) as the replacement for whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize. Checked 2026-11-15.
OpenAI gpt-4o-transcribeNot statedNot statedNot statedDeprecated. OpenAI's pricing page marks it deprecated; its deprecations page gives notice date 2026-08-26, shutdown 2027-02-26, replacement gpt-transcribe (files) or gpt-live-transcribe (live). Checked 2026-11-15.
OpenAI gpt-4o-mini-transcribeNot statedNot statedNot statedDeprecated. OpenAI's pricing page marks it deprecated; its deprecations page gives notice date 2026-08-26, shutdown 2027-02-26, replacement gpt-transcribe (files) or gpt-live-transcribe (live). Checked 2026-11-15.
OpenAI whisper-1Not statedNot statedNot statedDeprecated. OpenAI's pricing page marks it deprecated; its deprecations page gives notice date 2026-08-26, shutdown 2027-02-26, replacement gpt-transcribe (files) or gpt-live-transcribe (live). Checked 2026-11-15.
Google Cloud Speech-to-Text V2, standard models (incl. Chirp)1 secondNot statedNo · The V2 table lists no free minutes (the V1 API lists 60 free minutes a month).Channels: each channel billed. Not stated on the pricing page. First 500,000 minutes a month; lower prices above that. Checked 2026-11-14.
Google Cloud Speech-to-Text V2, dynamic batch1 secondNot statedNoChannels: each channel billed. Dynamic batch "processes audio at a lower level of urgency" at a discounted rate. Checked 2026-11-14.
Amazon Transcribe1 secondnoneYes · 60 minutes a month for 12 months.Channels: two channels billed as one duration. The pricing page lists speaker diarization among features included in the standard rate. Standard tier price; lower tiers at higher monthly volume. Checked 2026-11-14.
Microsoft Azure AI Speech, standard (East US)1 secondNot statedYes · F0: 5 audio hours free per month of real-time transcription; batch is not supported on the free tier.The pricing page lists diarization under enhanced add-on features, priced per hour for real-time; the amount is not shown in the page text. Prices from Microsoft's own Retail Prices API, because the pricing page renders prices client-side. Checked 2026-11-14.
Deepgram Nova-3, monolingual1 secondNot statedYes · 200 USD free credit; no credit card required to start.Channels: each channel billed. Included on pre-recorded; an add-on on streaming (USD 0.0020 a minute). The speech-to-text section is headed "Promo pricing"; streaming shows a struck-through regular price. Checked 2026-11-14.
Deepgram Nova-3, multilingual1 secondNot statedYes · 200 USD free credit; no credit card required to start.Channels: each channel billed. Included on pre-recorded; an add-on on streaming (USD 0.0020 a minute). Checked 2026-11-14.
AssemblyAI Universal-2Not statedNot statedYes · 50 USD in free credits, no credit card required.Channels: each channel billed. +0.02 USD per hour async; +0.12 USD per hour streaming, charged on the whole session. Streaming is billed on WebSocket session duration, not on audio sent. Universal-Streaming Multilingual covers English, Spanish, German, French, Portuguese and Italian, not Arabic; Arabic applies to the Universal-2 batch price. Checked 2026-11-14.
AssemblyAI Universal-3.5 ProNot statedNot statedYes · 50 USD in free credits, no credit card required.Channels: each channel billed. +0.02 USD per hour async; +0.12 USD per hour streaming. Checked 2026-11-14.
Speechmatics Standard accuracy (Pro, pay as you go)1 secondNot statedYes · 100 USD in free credits, no card required; credits last 12 months."Included for free" in batch and real-time. Pay-as-you-go standard rate. The pricing page displays rates after its 33% model-training discount (you let Speechmatics train on your data) and the maximum 25% volume discount, e.g. 0.23 USD an hour for Standard and 0.38 for Enhanced batch. Checked 2026-11-14.
Speechmatics Enhanced accuracy (Pro, pay as you go)1 secondNot statedYes · 100 USD in free credits, no card required; credits last 12 months."Included for free" in batch and real-time. Pay-as-you-go standard rate. The pricing page displays rates after its 33% model-training discount (you let Speechmatics train on your data) and the maximum 25% volume discount, e.g. 0.23 USD an hour for Standard and 0.38 for Enhanced batch. Checked 2026-11-14.
Soniox (async file and real-time)Not statedNot statedNot stated · No free credit mentioned on the pricing page."Speaker diarization, language identification, and smart formatting are bundled into the hourly rate." Also priced per token (1.50 USD per 1M input audio tokens async, 2.00 USD real-time); the hourly figure is Soniox's own. Checked 2026-11-14.
Rev.ai Reverb Foreign Language Transcription1 second15 secondsYes · Free credits equivalent to 5 hours of Reverb ASR.Not stated on the pricing page. Rev.ai's English Reverb model is 0.20 USD per hour; it does not transcribe Arabic. Checked 2026-11-14.
Rev.ai Reverb Transcription (English)1 second15 secondsYes · Free credits equivalent to 5 hours of Reverb ASR. Checked 2026-11-14.
Gladia Solaria-1 (self-serve)Not statedNot statedYes · 50 EUR in free credits for new accounts, valid for one year.Speaker diarization listed for async only on self-serve. Priced in euros; kept out of the USD statistics. Rates drop with monthly volume. Checked 2026-11-14.
Gladia Solaria-3 (self-serve)Not statedNot statedYes · 50 EUR in free credits for new accounts, valid for one year.Speaker diarization listed for async only on self-serve. Priced in euros; kept out of the USD statistics. Checked 2026-11-14.
ElevenLabs Scribe v3 / v2Not statedNot statedYes · A Free plan with included Scribe hours is listed; we did not confirm which hour figure belongs to which plan.Not stated on the API pricing page. Keyterm prompting and transcript editing cost extra per hour. Checked 2026-11-14.
Groq whisper-large-v3Not stated10 secondsYes · Free plan: 20 requests a minute, 2,000 a day, 7,200 audio seconds an hour and 28,800 a day.Not mentioned. Checked 2026-11-14.
Groq whisper-large-v3-turboNot stated10 secondsYes · Free plan: 20 requests a minute, 2,000 a day, 7,200 audio seconds an hour and 28,800 a day.Not mentioned. Checked 2026-11-14.
DeepInfra openai/whisper-large-v3Not statedNot statedNot statedPrice from DeepInfra's model page; the main pricing page did not list Whisper on the check date. Checked 2026-11-14.
DeepInfra openai/whisper-large-v3-turboNot statedNot statedNot statedPrice from DeepInfra's model page; the main pricing page did not list Whisper on the check date. Checked 2026-11-14.
BroadNet Transcribe ArabicNot statedNot statedYes · 5 USD starting credit, no card needed; both transcription models run on it.Not described in the API docs. File transcription only; WAV or MP3 up to 25 MiB per file. Prices synced from pricing.json on 2026-11-14. Checked 2026-11-14.
BroadNet TranscribeNot statedNot statedYes · 5 USD starting credit, no card needed; both transcription models run on it.Not described in the API docs. File transcription only; WAV or MP3 up to 25 MiB per file. Prices synced from pricing.json on 2026-11-14. Checked 2026-11-14.

Method

  • Each third-party value was read with AI assistance from the provider's own pricing, documentation or price-list API page on 2026-11-14; the JSON keeps the URL, check date and a quote for every price. Nothing comes from OpenRouter or another aggregator.
  • Batch is the cheapest standard pre-recorded (file) option at pay-as-you-go list price, first volume tier, default region. Streaming is the provider's real-time price for the same product family.
  • Per-minute prices are multiplied by 60 to give a price per audio hour, rounded to four decimals. Promotional prices are used where the provider shows them as the current price, and flagged.
  • Statistics cover USD rows that are not deprecated. Deprecated OpenAI models and euro-priced Gladia rows are listed but left out of medians and ranks; no currency conversion is applied.
  • Arabic is "yes" only where the provider's own language page names Arabic for that model; "not stated" means we could not confirm it, not that it is unsupported.
  • BroadNet rows come from pricing.json (synced 2026-11-14) through src/data/product.ts, so they change when the rate card changes.

Left out:

  • Fireworks AI, Whisper V3 and Whisper V3 Turbo: No speech-to-text price on fireworks.ai/pricing, the Whisper model pages or the transcription docs on the check date.
  • OpenAI, gpt-4o-transcribe-diarize: Deprecated diarization variant at the same price as gpt-4o-transcribe (USD 0.006 a minute on OpenAI's pricing page); left out to avoid a duplicate row. OpenAI's deprecations page gives shutdown 2027-02-26, replacement gpt-live-transcribe or gpt-transcribe.
  • Rev.ai, Whisper Large (English only) and human transcription: English-only Whisper hosting and human transcription are outside the scope of this index.
  • DeepInfra, Voxtral and Nemotron speech models: Listed per minute on deepinfra.com/pricing, but outside the brief (Whisper only); may be added in a later version.

Update frequency: third-party prices are re-read with AI assistance and the check date changes when they are; BroadNet rows and every statistic are recomputed on each build. Limits: list prices only, no volume discounts, enterprise contracts, taxes or accuracy. Code: src/lib/dataset-stt-prices.ts.

What do the CSV columns mean?

ColumnMeaning
idRow identifier.
providerCompany selling the API.
modelModel or product as the provider names it.
currencyUSD, or EUR where the provider prices in euros (not converted).
currentfalse if the provider marks the model deprecated.
batch_per_hourPre-recorded (file) list price per audio hour.
batch_price_is_fromtrue if the provider gives a "starts at" price.
streaming_modelReal-time product the streaming price belongs to.
streaming_per_hourReal-time list price per audio hour (session hour for AssemblyAI); empty if none.
streaming_promotrue if the streaming price is shown as promotional.
streaming_regular_per_hourStruck-through regular streaming price, where shown.
streaming_premiumstreaming_per_hour ÷ batch_per_hour.
billing_incrementStated billing increment; empty if not stated.
minimum_chargeStated minimum billed length per request; "none" if the page says no minimum.
channelsHow multi-channel audio is billed, where stated.
arabicyes, no, or empty if not stated on the provider's pages.
arabic_dialectsArabic locales or dialects the provider lists, separated by "; ".
diarizationincluded, extra, or empty if not stated.
free_tieryes, no, or empty if not stated.
free_tier_noteFree credit or minutes as the provider states them.
noteVolume tiers, deprecation and other caveats.
source_urlsProvider pages each value was read from, separated by " ".
checkedDate the sources were read (YYYY-MM-DD).

How to cite this dataset

Our compilation (the selection, per-hour conversions, statistics, rankings, tables and method): CC BY 4.0. Free to use, share and adapt with attribution and a link to this page.

Each price, language list and policy is a fact published by the provider named in its row, at the source URL given per row; we do not license those pages to you. Credit the provider as the source of each price.

Plain text

BroadRouter (2026). Speech-to-text API price index: batch and streaming price per audio hour (version 2026-11-15) [Data set]. BroadNet Technologies. https://broadrouter.com/data/speech-to-text-api-prices/

BibTeX

@misc{broadrouter_stt_price_index,
  author       = {BroadRouter},
  title        = {Speech-to-text API price index: batch and streaming price per audio hour},
  year         = {2026},
  version      = {2026-11-15},
  publisher    = {BroadNet Technologies},
  howpublished = {\url{https://broadrouter.com/data/speech-to-text-api-prices/}},
  note         = {Data set, data as of 2026-11-15},
}

Questions

What is the cheapest speech-to-text API per hour?

In checks from 2026-11-14 to 2026-11-15, the lowest listed batch price was USD 0.012 per audio hour for DeepInfra openai/whisper-large-v3-turbo. List price is not the whole bill: check the billing increment, minimum charge, per-channel billing and whether your language is supported.

What is the cheapest speech-to-text API for Arabic?

Among products whose provider names Arabic support, Soniox (async file and real-time) listed the lowest batch price, USD 0.10 per hour. BroadNet Transcribe (BroadNet, this site's owner) was USD 0.12. Accuracy on your dialect matters more than a few cents: test with your own audio.

Is streaming transcription more expensive than batch?

Usually a little. The median streaming price was 1.12× the batch price for the 12 products that list both, and 3 charge the same. Some gaps are large: Microsoft Azure AI Speech, standard (East US) charges 5.56× for real time.

How much does speech-to-text cost per minute?

Divide the hourly price by 60. The median batch price of USD 0.20 per hour is about USD 0.003 per minute. Estimate a monthly bill with the speech-to-text cost calculator.

Does speaker diarization cost extra?

It depends on the provider. In this data, 6 products include it, 3 charge extra (AssemblyAI adds a per-hour fee; Deepgram charges for it on streaming only), and 12 do not say on their pricing pages.

Where do these prices come from?

From each provider's own pricing, documentation or price-list API page, read with AI assistance on the check date. Every row in the CSV and JSON carries its source URLs. Nothing is taken from OpenRouter or other aggregators. BroadNet's rows come from its published pricing.json.

Can I use this data?

Yes. Our compilation, conversions and statistics are CC BY 4.0; cite them as shown under "How to cite". The prices themselves are facts published by each provider: credit the provider, and re-check the source page before you rely on a price.

Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity