OpenAI's open-source Whisper speech-to-text model, served on Cloudflare's edge with a free daily allowance and per-audio-minute pricing.
Last verified · 2026-07-16 · by Moe Ameen
"Whisper on Cloudflare Workers AI" is OpenAI's open-source Whisper model — the MIT-licensed automatic speech recognition (ASR) model most transcription tools are built on — hosted as a ready-to-call endpoint on Cloudflare's edge network. Instead of downloading model weights and renting a GPU, you POST an audio file to a Workers AI model ID and get back a transcript. Cloudflare runs three Whisper variants: `@cf/openai/whisper` (the base model), `@cf/openai/whisper-tiny-en` (a small English-only version trained on 680k hours of labelled audio), and `@cf/openai/whisper-large-v3-turbo` (the faster, more accurate large model). It is not a consumer app with a record button — it's an inference endpoint for developers and automations.
Whisper itself is a general-purpose, multilingual ASR model: it transcribes speech, translates non-English speech into English, and identifies the spoken language automatically. On Cloudflare, the response is structured for captioning — you get the full transcript text, a word count, a word-level array with timings, and a ready-made `vtt` (WebVTT) subtitle track you can drop straight onto a video. The Turbo model adds voice-activity detection, beam-search decoding, and optional filters that reduce the "hallucinated" phantom lines Whisper can emit over silence.
The pull is price and reach. Base Whisper is listed at $0.00045 per audio minute and Turbo at $0.00051 per audio minute, and all Workers AI usage draws from a free daily allocation (10,000 Neurons per day, resetting at 00:00 UTC) before you pay anything — enough for a few hours of transcription a day at no cost. Because it runs on Cloudflare's edge, a Worker can transcribe audio as part of an automated pipeline without you managing any infrastructure.
The honest framing: this is a transcription endpoint, not a content tool. It converts audio into accurate text and a subtitle file, and it stops there. It writes no post, cuts no clip, designs no carousel, holds no brand voice, and publishes to nothing. It gives you the words — cheaply, at scale, with timestamps — and the rest of the job is still yours.
The cleanest way to see the split: Workers AI Whisper hands you a `.vtt` file and a transcript. That's an ingredient, not a deliverable — a subtitle track is one line on a video, and a transcript is a wall of text nobody watches. Kompozy is the engine that turns that raw material into a finished, published content week. Instead of stopping at "here are the captions," bring the recording itself into Kompozy: Clipped Shorts finds the moments worth posting inside your long video and burns in styled captions automatically (so the cheap edge transcript becomes an actual short, not just an overlay file), while from the transcript text Kompozy drafts a Blog Article and an Email Newsletter, pulls the sharpest lines into Quote Graphics and a brand-exact Carousel, and writes native Text Posts — every piece governed by your Persona Brief so the whole batch reads in one voice instead of verbatim dictation.
The multiplier is that one recording becomes 25–35 outputs and then goes everywhere. Whisper on Cloudflare stops at the words and their timestamps; Kompozy reframes to 9:16, 1:1, and 16:9 and fans a single source across nine social platforms plus a blog and email newsletter from one queue, with Autopilot and a per-post review pipeline. It also generates formats a transcription endpoint physically can't produce — Persona Shorts and HeyGen avatar video with a face-locked recurring identity, Persona Frames, and Marketing Shorts — so the same idea can also ship as a talking-head clip. Use Workers AI Whisper to get accurate, timestamped words for a fraction of a cent a minute; use Kompozy to turn those words into finished, scheduled, on-brand posts across every platform.
No. Cloudflare Workers AI hosts OpenAI's actual open-source Whisper model (MIT-licensed) as a callable endpoint. It runs three variants — the base `@cf/openai/whisper`, the English-only `whisper-tiny-en`, and the faster `whisper-large-v3-turbo`. It's the same model most transcription tools use, just served on Cloudflare's edge with a free daily allowance.
Base Whisper is listed at $0.00045 per audio minute and Whisper Large V3 Turbo at $0.00051 per audio minute. All Workers AI usage draws from a free daily allocation of 10,000 Neurons (resetting at 00:00 UTC) before any charge, which covers a few hours of transcription a day at no cost. Confirm current rates on Cloudflare's Workers AI pricing page.
Yes. The response includes a WebVTT (`vtt`) subtitle track with timings alongside the plain transcript text and word count, so you can drop captions onto a video directly. The Turbo model also adds voice-activity detection and hallucination filtering to keep the timing and text cleaner.
No. It transcribes audio into text and a subtitle file and nothing more — it writes no posts, cuts no clips, designs no carousels, and publishes nowhere. To turn a transcript into captioned shorts, carousels, a blog, a newsletter, and scheduled posts across platforms, you use a content engine like Kompozy.
Transcribe the recording through Workers AI, then bring the recording and transcript into Kompozy. Kompozy cuts captioned Clipped Shorts from the audio, drafts a blog and newsletter from the transcript, builds carousels and quote graphics from the best lines in your Persona Brief voice, and schedules and publishes them across nine platforms plus blog and email.