// AI NEWS · FEATURE

Whisper-Grade Transcription Now Runs on Cloudflare's Edge for a Fraction of a Cent per Minute

Cloudflare Workers AI serves OpenAI's open-source Whisper — including the faster Large V3 Turbo model — as a hosted endpoint for roughly $0.0005 per audio minute, inside a free daily allowance. Accurate speech-to-text has quietly become an edge commodity.

KompozyTurn one idea into a week of content — across every platform, published for you.
Get Started →

2026-07-16 · by Moe Ameen

What happened

Cloudflare's Workers AI platform hosts OpenAI's open-source Whisper — the MIT-licensed speech-to-text model most transcription tools are built on — as a set of ready-to-call edge endpoints. Alongside the base `@cf/openai/whisper` and the English-only `whisper-tiny-en`, Cloudflare offers `whisper-large-v3-turbo`, the faster, more accurate large model, which graduated out of beta with per-audio-minute pricing on March 17, 2025 as part of a batch of new Workers AI models. There's no single new launch here so much as a state of play worth naming: high-quality, multilingual transcription now runs on a global edge network as a metered API, not a GPU you have to rent.

The pricing is the story. As verified on Cloudflare's Workers AI documentation, base Whisper is listed at $0.00045 per audio minute and Turbo at $0.00051 per audio minute, and all Workers AI usage — Whisper included — draws from a free daily allowance of 10,000 Neurons (resetting at 00:00 UTC) before any charge, with paid usage billed at $0.011 per 1,000 Neurons. In practical terms, a creator or developer can transcribe a few hours of audio a day at no cost and a large backlog for a handful of dollars.

The output is built for reuse. A call returns the full transcript text, a word count, a word-level timing array, and a ready-made WebVTT subtitle track; the Turbo model adds voice-activity detection and hallucination filtering to keep long-form transcripts clean. What it does not do is anything past the words: it writes no post, cuts no clip, designs no carousel, and publishes nothing. It's an endpoint that turns audio into accurate, timestamped text at the edge — cheaply, at scale — and stops there.

Why it matters for creators

  • Transcription cost is collapsing toward zero. At a fraction of a cent per minute with a free daily tier, "how do I afford to transcribe everything?" stops being a real question — captioning and transcript-mining are now effectively free at a creator's scale.
  • Your entire back-catalog becomes cheap to unlock. Years of podcasts, interviews, and livestreams can be transcribed in bulk for a few dollars, turning a dormant archive into searchable, repurposable source text.
  • It runs on the edge as an API, not an app. That makes automated captioning pipelines trivial for developers — but a non-developer still can't use the endpoint directly, and it returns only text and a .vtt.
  • The bottleneck moves downstream. When getting the words out of audio is instant and cheap, the scarce work is everything after: turning those words into clips, posts, and published content in your voice.
  • Cheap transcription is not cheap content. The endpoint stops at a transcript and a subtitle file — no clip, no carousel, no post, no schedule — so the value now lives in what you do with the words, not in getting them.

How to act on this with Kompozy

The interesting consequence of near-free edge transcription isn't the caption file — it's what it does to your archive. Transcribe a year of podcasts or livestreams through Workers AI Whisper for a few dollars and you suddenly have a searchable library of every idea you've ever said out loud. That's exactly the raw material Kompozy is built to mine. Feed those recordings in and Kompozy's Clipped Shorts hunts the moments worth posting across the whole back-catalog and burns in styled captions; from the transcripts, it drafts Blog Articles and Email Newsletters, pulls the sharpest lines into Quote Graphics and brand-exact Carousels, and writes native Text Posts — every piece held to one voice by your Persona Brief so a mined archive reads as your brand, not a pile of dictation.

That turns a one-time transcription spend into a repeatable content engine. Workers AI Whisper stops at timestamped words; Kompozy reframes them to 9:16, 1:1, and 16:9 and fans a single recording into 25–35 outputs across nine social platforms plus blog and email from one queue, with Autopilot and a per-post review pipeline — and generates the formats a transcription endpoint can't touch at all, like Persona Shorts and HeyGen avatar video with a face-locked recurring identity. There's a same-week publishing angle too: "transcription is now an edge commodity" is a shift your audience is curious about, and Kompozy turns your take on it into a captioned short, a carousel, a blog explainer, and platform-native posts in an afternoon.

Quick takeaways

  • Cloudflare Workers AI hosts OpenAI's open-source Whisper (base, tiny-en, and large-v3-turbo) as edge endpoints; Turbo graduated out of beta with pricing on March 17, 2025.
  • Pricing as verified July 2026: base Whisper ~$0.00045/audio min, Turbo ~$0.00051/audio min, inside a free daily allowance of 10,000 Neurons (paid usage $0.011/1,000 Neurons).
  • The response includes transcript text, word-level timings, and a ready WebVTT subtitle track; Turbo adds voice-activity detection and hallucination filtering.
  • It transcribes and stops — no clipping, content generation, or publishing. Kompozy turns the transcript into finished, scheduled posts across nine platforms plus blog and email.

Frequently asked questions

How much does Whisper cost on Cloudflare Workers AI?

As listed in Cloudflare's Workers AI documentation, base Whisper is $0.00045 per audio minute and Whisper Large V3 Turbo is $0.00051 per audio minute. All usage draws from a free daily allowance of 10,000 Neurons (resetting at 00:00 UTC), with paid usage billed at $0.011 per 1,000 Neurons. Confirm current rates on Cloudflare's pricing page.

Is Cloudflare Whisper free?

Partly. All Workers AI usage, including Whisper, draws from a free daily allocation of 10,000 Neurons that resets at 00:00 UTC — enough for a few hours of transcription a day at no cost. Heavier use is metered per audio minute beyond the free tier.

Can I turn Cloudflare Whisper transcripts into social content?

Not with the endpoint itself — it returns text, timings, and a subtitle file and nothing more. To turn a transcript into captioned Clipped Shorts, carousels, quote graphics, a blog, and a newsletter, and to schedule and publish them across nine platforms, you bring the recording into a content engine like Kompozy.

Related news

← All AI news · Get started →