> ## Documentation Index
> Fetch the complete documentation index at: https://hyperwhisper.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Providers
> Pick a transcription model by language accuracy and cost. Covers HyperWhisper Cloud tiers, silence-free billing, and BYOK alternatives.
HyperWhisper supports four transcription tiers out of the box via HyperWhisper Cloud, plus bring-your-own-key (BYOK) for direct provider access and fully offline local models.
## HyperWhisper Cloud at a glance
HyperWhisper Cloud is built-in — no API key, no separate account. Pick a tier based on whether you care more about speed, balance, or accuracy. All four are pay-as-you-go with no markup; you pay what the underlying provider charges.
Our top-accuracy tier. Best results on accents, noisy environments, and technical vocabulary.
**\~\$0.59 / hour** · *9.83 credits/min*
**Deepgram Nova-3**
\~\$0.33 / hour
*5.5 credits/min*
Strong English accuracy, low latency, custom vocabulary.
**Grok STT (xAI)**
\$0.10 / hour
*1.67 credits/min*
Solid multilingual accuracy at a low per-minute cost.
**Groq Whisper Large v3**
\~\$0.11 / hour
*1.85 credits/min*
Sub-second latency. Great for English and major European languages.
Credits are billed at **1 credit = \$0.001 USD**. A Pro license includes 5,000 credits up front, and top-ups are available in \$5 / \$10 / \$20 bundles.
## You only pay for actual speech
HyperWhisper Cloud detects silence and blank audio automatically. If a recording contains no detectable speech, **you are charged 0 credits** — we don't bill for dead air at the start of a clip, pauses between thoughts, or an accidentally-triggered empty recording.
In practice, across a typical working day of push-to-talk dictation, you're only billed for the minutes you actually spoke.
## Accuracy by language
For English, any tier works. For the best HyperWhisper Cloud quality, use **Highest (ElevenLabs Scribe v2)** — the top accuracy tier, especially on accents, noisy audio, and technical vocabulary. Most users find **High (Deepgram Nova-3)** is plenty for everyday dictation, and the **Medium / Fast** tiers (xAI Grok, Groq Whisper) are great when cost and latency matter more than the last few percent of accuracy.
xAI does not publish a per-language WER table for Grok STT. We avoid mixing older third-party benchmark numbers with this tier because they do not measure the same provider.
## Cost examples
At 1 credit = \$0.001 USD, here's what each tier costs at typical usage levels. Remember: **only actual speech is billed**, so "30 min/day" means 30 minutes of talking, not 30 minutes of the app being open.
| Daily speech | Highest (ElevenLabs) | High (Deepgram) | Medium (Grok) | Fast (Groq) |
| ------------ | -------------------- | --------------- | ------------- | ----------- |
| 15 min | \~\$0.15 | \~\$0.08 | \~\$0.03 | \~\$0.03 |
| 30 min | \~\$0.30 | \~\$0.17 | \~\$0.05 | \~\$0.06 |
| 1 hour | \~\$0.59 | \~\$0.33 | \~\$0.10 | \~\$0.11 |
| 2 hours | \~\$1.18 | \~\$0.66 | \~\$0.20 | \~\$0.22 |
| 8 hours | \~\$4.72 | \~\$2.64 | \~\$0.80 | \~\$0.89 |
Monthly at 30 min/day: **\~\$9 Highest / \~\$5 High / \~\$1.50 Medium / \~\$1.80 Fast.** A one-time \$5 top-up covers roughly 8 hours of speech on the Highest tier or \~50 hours on the Medium tier.
## Alternatives
If you already have API credits or want to use your own free tier (Deepgram \$200, AssemblyAI \$50), plug in a key via [API Keys](/docs/api-keys). You pay the provider directly at their published rate.
| Provider | Model | \$/min |
| ---------- | ----------------------------- | --------- |
| Groq | Whisper Large v3 Turbo | \$0.00067 |
| xAI | Grok STT | \$0.00167 |
| Deepgram | Nova-3 (batch) | \$0.0043 |
| AssemblyAI | Universal | \$0.0037 |
| OpenAI | whisper-1 / gpt-4o-transcribe | \$0.006 |
| ElevenLabs | Scribe v2 | \~\$0.008 |
HyperWhisper also supports Mistral and Google Gemini for BYOK. See [API Keys](/docs/api-keys) for setup.
When you bring your own key, opting your audio out of model training is your responsibility — every provider has its own dashboard setting. See [Data Privacy & Model Training](/docs/data-privacy) for a copy-pasteable LLM prompt that finds the current opt-out for any provider.
Fully offline models (Whisper, Parakeet, Nemotron) run on-device with no network and no cost. Accuracy is slightly lower than the best cloud tier, but perfectly usable for English and larger European languages. See [Models](/docs/models).
## Boost accuracy on any provider
* **[Custom vocabulary](/docs/vocabulary)** — add domain terms (product names, frameworks, jargon, colleagues' names). Biggest single improvement for technical or professional use.
* **Low-noise environment** — every model degrades with background noise. See [Best Practices](/docs/best-practices).
* **Natural pace** — overly fast or overly slow speech both hurt accuracy.
When using Deepgram Nova-3 with custom vocabulary, set the language explicitly (not `auto`) — the `keyterm` parameter is only active in monolingual mode on Nova-3.
## FAQ
**Does HyperWhisper Cloud mark up the underlying provider cost?**
No. You pay the same per-minute rate as if you held the provider's API key directly.
**What happens if my chosen tier is temporarily unavailable?**
HyperWhisper Cloud automatically falls back to another provider in the chain, so transcription still succeeds. You're billed at the actual provider that handled the request.
**Which tier should I use for the highest quality?**
Highest (ElevenLabs Scribe v2) is the top HyperWhisper Cloud tier — best accuracy on accents, noisy audio, and technical vocabulary. High (Deepgram Nova-3) is a strong, lower-cost default for most everyday dictation.