ElevenLabs Scribe is one of the stronger speech-to-text APIs: 90+ languages, speaker diarization, keyterm prompting and real-time models, from the same company as ElevenLabs text to speech. This page is for people with a specific reason to look elsewhere, whether that is API format, data retention, price or Arabic. It compares six alternatives, including BroadNet Transcribe, and says plainly when Scribe is still the better choice. It covers speech to text only, not ElevenLabs voices.
What does ElevenLabs Scribe do well?
From ElevenLabs’ own pages, checked 2026-11-15.
- Recent, accurate models. Scribe v3 is its “most accurate speech recognition model for recorded audio”. ElevenLabs reports it cuts WER on FLEURS across 102 languages from 16.94% (Scribe v2) to 12.38%.
- Real time. Scribe v3 Live and Scribe v2 Realtime stream transcripts; ElevenLabs gives Scribe v2 Realtime a median of about 150 ms to a committed transcript.
- Many languages. 90+ languages on Scribe v3 and v2.
- Features. Diarization for up to 32 speakers, word-level timestamps, audio events such as laughter, entity detection, and up to 1,000 keyterms on batch.
- Large files. Up to 5 GB and 15 hours per file, in many audio and video formats, including OGG, M4A and MP4.
- Concurrency. Batch speech-to-text limits run from 8 on Free to 40 on Pro and 60 on Scale and Business.
- One vendor for voice. If you also use ElevenLabs text to speech or agents, one account covers both.
Why do people look for an ElevenLabs Scribe alternative?
- Zero retention needs an enterprise contract. The API reference says zero retention mode “may only be used by enterprise customers”. Without it, ElevenLabs keeps history for speech-to-text requests by default.
- Its own API format. Scribe uses
POST https://api.elevenlabs.io/v1/speech-to-textwith anxi-api-keyheader. Tools built for the OpenAI-compatible transcription endpoint need a separate integration. - Price for bulk files. USD 0.20 an hour is more than Groq Whisper (USD 0.04 to USD 0.111), Soniox (USD 0.10), AssemblyAI Universal-2 (USD 0.15) and BroadNet Transcribe ($0.12). Keyterm prompting adds USD 0.05 an hour on batch. Model your volume with the speech-to-text cost calculator.
- Arabic dialects. ElevenLabs puts Arabic in its “Good” accuracy group for Scribe v2, meaning over 10% and up to 20% word error rate. That is solid, but Gulf, Egyptian and Levantine audio deserves a test against a dedicated Arabic model.
ElevenLabs Scribe vs alternatives: comparison table
| ElevenLabs Scribe | Groq | AssemblyAI | OpenAI | BroadNet Transcribe | |
|---|---|---|---|---|---|
| Example batch price | Scribe v3 USD 0.20/hr | Whisper large-v3-turbo USD 0.04/hr | Universal-2 USD 0.15/hr | gpt-transcribe USD 0.0045/min; deprecated gpt-4o-mini-transcribe USD 0.003/min | $0.002/min ($0.12/hr); Arabic $0.0042/min ($0.25/hr) |
| Real time | Yes, USD 0.22/hr | Not listed | Yes, from USD 0.15/hr | Realtime models, priced separately | No, files only |
| OpenAI-compatible | No (/v1/speech-to-text) | Yes | No for STT | The reference | Yes |
| Data retention (cited) | History kept by default; zero retention mode for enterprise only | Not retained by default; logs up to 30 days unless ZDR, open to all | Audio deleted within 24 to 48 hours; transcripts from 30 days; TTL configurable | No abuse-log retention on /v1/audio/transcriptions | “keeps no prompts, messages, images, audio, responses or transcripts” |
| Languages | 90+ | Whisper’s languages | 99 (Universal-2) | Many | 14, plus a dedicated Arabic model |
| Diarization | Yes, up to 32 speakers | Not listed | +USD 0.02/hr | Not compared here | Not listed |
| File limit | 5 GB, 15 hours | 25 MB free, 100 MB dev tier (larger by URL) | Not compared here | Not compared here | 25 MiB, WAV or MP3 |
Sources for every cell are listed below. For more providers, see speech-to-text API pricing compared.
What does BroadNet’s own Arabic benchmark show?
BroadNet publishes a head-to-head on its Arabic model page. These are BroadNet’s own figures on its own benchmark, word error rate, lower is better:
| Test set (BroadNet’s benchmark) | BroadNet Transcribe Arabic | ElevenLabs Scribe v2 | OpenAI Whisper large-v3 |
|---|---|---|---|
| MASC, real dialect speech | 21.1 | 22.3 | 31.5 |
| Gulf, Egyptian and Levantine set | 23.4 | 25.6 | 31.8 |
Read this with care. The gap to Scribe v2 is small. BroadNet does not publish its dataset size or normalisation rules there. The comparison is against Scribe v2, not the newer Scribe v3, which ElevenLabs says is more accurate. Run both on 30 to 50 of your own clips before deciding. Arabic WER explained shows how to normalise Arabic text first.
6 alternatives to ElevenLabs Scribe at a glance
Batch (file) list prices, checked 2026-11-15, lowest US-dollar price first, then BroadNet’s own product.
- Groq (Whisper large-v3-turbo): USD 0.04 an hour, OpenAI-compatible, billed at least 10 seconds per request. The lowest price we found.
- Soniox: USD 0.10 an hour for files and USD 0.12 for real time, 60+ languages including Arabic.
- AssemblyAI (Universal-2): USD 0.15 an hour, diarization for USD 0.02 an hour more, and USD 50 of free credit with no card.
- OpenAI: gpt-transcribe is USD 0.0045 a minute (USD 0.27 an hour). The older gpt-4o-mini-transcribe (USD 0.003 a minute, USD 0.18 an hour), gpt-4o-transcribe and whisper-1 (USD 0.006) are marked deprecated on its pricing page.
- Deepgram (Nova-3): USD 0.0043 a minute monolingual (about USD 0.26 an hour), more than Scribe, but with USD 200 of free credit and strong streaming. See our Deepgram alternative page.
- BroadNet Transcribe: $0.12 an hour for 14 languages and $0.25 for the dedicated Arabic model, OpenAI-compatible, with no audio or transcripts kept. Note that the Arabic model costs more per hour than Scribe. This is BroadNet’s own product.
When should you stay with ElevenLabs Scribe?
- You need many languages. 90+ languages, against 14 on BroadNet Transcribe (Arabic among them) plus the dedicated Arabic model.
- You need diarization, entity detection or keyterms. Scribe has them built in; BroadNet lists none of these.
- You stream live audio. Scribe v3 Live and v2 Realtime cover real-time captions and voice bots. BroadNet does file transcription only.
- Your files are large or not WAV/MP3. Scribe takes up to 5 GB and video formats directly.
- You already use ElevenLabs voices and want one vendor and one bill.
- You are on an enterprise contract that already includes zero retention mode.
When does BroadNet Transcribe fit?
BroadNet Transcribe runs on BroadNet Inference; BroadNet’s privacy statement says that runs on BroadNet-owned or leased hardware in the United States, with partner capacity as backup. BroadNet’s model pages list NVIDIA GPUs it operates in the United States for both speech models.
- Arabic is your main language. BroadNet Transcribe Arabic is a dedicated Arabic model. BroadNet reports it ranked #2 of 37 (its September 2026 figure; the live board listed it #4 of 39 on 15 November 2026) on the Open Universal Arabic ASR Leaderboard. Its own benchmark above puts it slightly ahead of Scribe v2. Both are BroadNet’s reports, so test it. See the Arabic speech-to-text guide.
- You want zero retention without an enterprise deal. BroadNet “keeps no prompts, messages, images, audio, responses or transcripts” on every account. See zero data retention.
- You use the OpenAI SDK. BroadNet Transcribe covers 14 languages on the same
/v1/audio/transcriptionscall, at $0.12 an hour. - You want an LLM on the same key to summarise or classify transcripts.
Limits to know: files are WAV or MP3 up to 25 MiB each, so convert OGG voice notes and split long recordings. BroadNet documents file transcription only, with JSON text output; it does not list streaming, diarization or timestamps. The Arabic model is $0.25 an hour, which is more than Scribe’s USD 0.20. The free credit is $5, about 1,190 minutes of Arabic.
When does another alternative fit better?
- Groq for the lowest-cost Whisper: USD 0.04 an hour, OpenAI-compatible, with a 10-second minimum per request.
- Soniox for low-cost files and streaming in 60+ languages, Arabic included. Add it to your Arabic test.
- AssemblyAI for 99 languages, cheap diarization and a USD 50 credit.
- OpenAI if you already use its API and want one bill.
Our Whisper API alternative post covers the Whisper-based options in more depth.
How do you switch from ElevenLabs Scribe?
Scribe’s /v1/speech-to-text call becomes an OpenAI-style transcription call:
from openai import OpenAI
import os
client = OpenAI(base_url="https://inference.broadnet.ai/v1", api_key=os.environ["BROADNET_API_KEY"])
# ElevenLabs: POST https://api.elevenlabs.io/v1/speech-to-text (model_id=scribe_v3)
with open("voice-note.mp3", "rb") as f:
result = client.audio.transcriptions.create(
model="broadnet/cohere-transcribe-arabic", # or broadnet/cohere-transcribe for 14 languages
file=f,
language="ar",
)
print(result.text)
- Pick 30 to 50 real clips per language, with human transcripts.
- Run them through Scribe v3 and each alternative, and compare word error rate.
- Convert other formats to WAV or MP3 and split files over 25 MiB on silence, so no word is cut at a boundary.
- Move batch work first. Keep Scribe for streaming, diarization or languages BroadNet doesn’t cover.
The short version
ElevenLabs Scribe is a strong choice for many languages, diarization, real time and teams already on ElevenLabs voices. If you need an OpenAI-compatible API, zero retention without an enterprise contract, or a lower per-hour price, test alternatives on your own audio. Groq is the low-cost Whisper option. BroadNet Transcribe fits Arabic-heavy, privacy-sensitive batch work, though its Arabic model costs more per hour than Scribe. For WhatsApp audio, see transcribing voice notes.