---
title: "GroqCloud (Whisper Large-v3)"
type: "AI Tool"
url: "https://aidemos.com/tools/groqcloud-whisper-large-v3"
description: "We transcribed a medical-jargon file at 3.15% WER and a bilingual sample at 27.94% WER, but a 65.39 MB file hit HTTP 413."
category: "audio-speech"
website: "https://console.groq.com/docs/speech-to-text"
published: "2026-08-13T09:18:22.557946+00:00"
updated: "2026-08-20T12:47:11.171825+00:00"
---

# GroqCloud (Whisper Large-v3)

Fast batch transcription for under-cap audio, with strong jargon handling but a hard 25 MB ceiling.

## TL;DR Verdict

**Strong on jargon, weaker on code-switching, blocked by oversize uploads.**

**Where it wins:**
- You need inexpensive batch transcription for audio under the documented 25 MB cap.
- You want strong technical-jargon handling on single-speaker narration.
- You can work with verbose JSON and segment-level timing but do not need native diarization.

**Main limitation:** Your audio files often exceed 25 MB.

**Pricing:** Whisper V3 Large (pay-as-you-go) $0.111 / audio hour · Whisper Large v3 Turbo (pay-as-you-go) $0.04 / audio hour · Free plan — Whisper V3 Large $0 · Free plan — Whisper Large v3 Turbo $0

`25 MB free-tier cap` · `3.15% WER on jargon` · `Word timestamps` · `No native diarization`

**Website:** [Visit GroqCloud (Whisper Large-v3)](https://console.groq.com/docs/speech-to-text)

> **Strong on jargon, weaker on code-switching, blocked by oversize uploads.**
>
> GroqCloud (Whisper Large-v3) is a strong low-cost batch STT engine for under-cap audio: it scored 3.15% WER on the medical-jargon file, recalled all 9 scored jargon terms, and ran very fast at 3.82s wall clock. It also produced usable output on the bilingual Spanish-English sample, but accuracy dropped sharply there (27.94% WER, 53.8% Spanish recall). The operational limit is the main risk: a 65.39 MB crosstalk file was rejected with HTTP 413 before transcription, and the service exposes no native speaker diarization.

## Demo Recording

[Video: GroqCloud (Whisper Large-v3) demo recording](https://cdn.futuresmart.ai/public/aidemos/b12b4bbff1e4482b9993b736429e703b.mov?v=1)
*Video — Screen recording of the Groq API keys page followed by a terminal benchmark run, ending with the oversized request rejection and benchmark selection flow.*

## Feature-by-Feature Breakdown

### Audio Transcription

**Verdict:** Works well on under-cap audio, but the 65.39 MB overlap file was rejected before transcription and the bilingual case was materially weaker than the medical-jargon run.

Groq’s audio transcription endpoint converts uploaded audio files into text transcripts, and the benchmark exercised it on an 8.58 MB medical-jargon file and a 22.19 MB bilingual file. The same endpoint was also tested with verbose JSON and word timestamps enabled, producing structured transcript outputs with timing and metadata.

**Input:**

> **Audio**

**Output:** Raw API response

> **Json** — Raw API response

**Input:**

> **Audio**

**Output:** Raw API response

> **Json** — Raw API response

**Input:** Same medical-jargon run metrics

> **Audio** — Same medical-jargon run metrics

**Output:** Run metrics

> **Image** — Run metrics

**Input:** Same bilingual run metrics

> **Audio** — Same bilingual run metrics

**Output:** Run metrics

> **Image** — Run metrics

**Input:**

> **Audio**

**Output:** Limit / error evidence

> **Image** — Limit / error evidence

**Bottom line:** Strong under the file-size cap and excellent on the technical-jargon input, but code-switching accuracy drops on harder bilingual audio and oversized files fail at the API boundary.

### Audio Upload Size Validation

**Verdict:** Clear and fast failure mode, but it blocks evaluation of the crosstalk scenario on this oversized file.

The API rejects oversized audio uploads at the boundary instead of attempting transcription. In the tested case, an over-limit file failed immediately with HTTP 413.

**Input:**

> **Audio**

**Output:**

**Bottom line:** Clear and fast guardrail, but the intended overlap test could not be scored because the upload exceeded the cap.

## Official Groq ASR pricing

Per-hour list prices and free-tier limits from Groq's speech-to-text docs.

| Plan | Price | Notes |
| --- | --- | --- |
| Whisper V3 Large (pay-as-you-go) ★ (tested) | $0.111 / audio hour | 100 MB max file (dev tier); 189x real-time speed factor; 99+ languages |
| Whisper Large v3 Turbo (pay-as-you-go) | $0.04 / audio hour | 100 MB max file (dev tier); 216x speed factor; transcription only (no translation) |
| Free plan — Whisper V3 Large | $0 | 20 RPM; 2,000 requests/day; 7,200 audio-sec/hour; 28,800 audio-sec/day; 25 MB max file |
| Free plan — Whisper Large v3 Turbo | $0 | Same free-tier limits as Whisper V3 Large |

*Free tier is capped at 25 MB per file; audio is billed at a 10-second minimum per request. Groq also documents a developer plan and Batch/Flex processing, but no STT-specific batch discount was published in the fetched docs.*

## Is It Right For You?

**Use it if**
- You need inexpensive batch transcription for audio under the documented 25 MB cap.
- You want strong technical-jargon handling on single-speaker narration.
- You can work with verbose JSON and segment-level timing but do not need native diarization.
- You want simple multipart POST integration with no extra routing setup.

**Skip it if**
- Your audio files often exceed 25 MB.
- You need native speaker diarization.
- You need word-accurate captions or search without a separate forced-alignment step.
- You need especially strong code-switching on Spanish-heavy spans.

## Classification

- **Category:** audio-speech
- **Subcategory:** other-audio-speech
- **Type:** speech
- **Built for:** Other

## Frequently Asked Questions

**Q: Does GroqCloud (Whisper Large-v3) have a file upload limit?**

Yes. In this benchmark, the 65.39 MB crosstalk file was rejected with HTTP 413, and the report notes a 25 MB free-tier cap. The overlap file never reached transcription, so its accuracy is untested here.

**Q: How accurate was it on technical jargon?**

Very strong on the medical-jargon input: 3.15% WER, 58 substitutions, 13 deletions, 15 insertions, and 100% recall of the 9 scored jargon terms.

**Q: How did it handle Spanish-English code-switching?**

It produced a transcript, but accuracy dropped on the bilingual input: 27.94% WER and 53.8% Spanish token recall. That makes it usable, but clearly weaker than the jargon case.

**Q: Does it provide speaker diarization?**

No native speaker labels were detected in the raw API response, and the report explicitly notes that Groq does not provide native speaker diarization here.

**Q: Does it return word-level timestamps?**

Word timestamps are enabled in the request and the raw response reports word_timestamps as present, but the benchmark only observed segment-level timing. The report says word-accurate captions or search would still need a separate forced-alignment pass.

**Q: What did it cost in the benchmark?**

The medical-jargon run cost $0.03465, the bilingual run cost $0.05977, and the rejected oversized request cost $0.0. The report's listed model price is $0.00185/min, or $0.111 per audio hour.

**Q: What endpoint and format did the benchmark use?**

A multipart POST to https://api.groq.com/openai/v1/audio/transcriptions with response_format set to verbose_json and word timestamps enabled.

## Similar Tools

AI tools similar to GroqCloud (Whisper Large-v3):

- [Deepgram Nova-3](https://aidemos.com/tools/deepgram-nova-3) — Batch speech-to-text with rich metadata, but weak on crosstalk and code-switching.
- [ElevenLabs Scribe](https://aidemos.com/tools/elevenlabs-scribe) — Fast batch transcription with rich metadata, strongest on jargon and weaker on overlap and code-switching.
- [Rev AI](https://aidemos.com/tools/rev-ai) — Low-cost batch speech-to-text with structured JSON, word timestamps, and speaker labels, but mixed accuracy on hard audio and weak multilingual recall.
- [Gladia](https://aidemos.com/tools/gladia) — Batch STT with rich JSON metadata and strong jargon recall, but weak overlap handling and a channel-duplication caveat on bilingual audio.
- [Speechmatics](https://aidemos.com/tools/speechmatics) — Strong batch STT for hard English audio, but weak on code-switching as configured.
- [ElevenLabs](https://aidemos.com/tools/elevenlabs) — Natural-sounding voice cloning and narration, but with only approximate voice identity.

## Need a custom AI solution for this use case?

If you are looking to build a custom speech-to-text, audio transcription, or batch transcription workflow for your business or internal workflow, email us at [contact@futuresmart.ai](mailto:contact@futuresmart.ai).

### Found something inaccurate or missing?

We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at [collaborate@aidemos.com](mailto:collaborate@aidemos.com).