Cloudflare Workers AI hosts OpenAI Whisper models for speech recognition on the edge. Public unit prices on the model pages (checked October 26, 2026) are $0.000453 per audio minute for @cf/openai/whisper and $0.000513 per audio minute for @cf/openai/whisper-large-v3-turbo (Batch). The Workers AI pricing audio table rounds both to about $0.0005 per audio minute.
The LecSync Realtime Speech API is a managed WebSocket for meeting and lecture captions: preview and final text, translation between any two of 60 languages, a speaker and audio-aligned timestamps on every final segment, and running minutes. It costs $0.006 per audio minute ($0.36 per hour) with those features included.
Figures below were checked on October 26, 2026 against Cloudflare's @cf/openai/whisper and whisper-large-v3-turbo model pages, the Workers AI pricing and limits docs, and the AI Gateway Realtime WebSockets API.
What is Cloudflare Workers AI Whisper vs a realtime STT caption API?
Cloudflare Workers AI Whisper is a batch / inference automatic speech recognition path: you send audio to a Workers binding or REST inference call and get a transcript (and, on the turbo model, optional task=translate into English plus VAD and decoding knobs). A realtime STT caption API such as LecSync keeps a WebSocket open, streams PCM as people talk, and returns paired transcripts, translations, speakers and minutes without waiting for a finished file.
The short answer
If you need cheap file or edge Whisper transcription inside the Cloudflare Workers ecosystem, Workers AI undercuts almost every managed caption bill. At model-page rates, @cf/openai/whisper is about $0.027 per hour of audio; turbo is about $0.031 per hour — an order of magnitude below LecSync's $0.36 per hour all-in.
If you need live translated captions, speakers and minutes on one flat minute bill while the session is still running, LecSync is the closer product. Whisper on Workers AI is not the Cloudflare realtime STT path; live streaming speech-to-text on AI Gateway lists Deepgram (Nova-3 / Flux) on Workers AI, not Whisper.
How each API works
LecSync. Your server calls POST https://api.lecsync.com/v1/realtime/connect with the audio format, a language pair and the features you want, then streams 16 kHz mono PCM. Preview text arrives while someone is still talking; a final segment settles once the sentence confirms. Translations share the segment ID. With speakers: true, each final carries a speaker and startMs/endMs on the recording clock. With digest: true, minutes arrive section by section. The first word shows up about 0.28 seconds after speech starts; translation adds roughly 88 ms. Seven gateway regions (dal, jp3, kc, nj, ru2, ru3, uk) keep audio near the capture point; backlog is capped at 2 seconds.
Cloudflare Workers AI Whisper. You run @cf/openai/whisper or @cf/openai/whisper-large-v3-turbo through a Workers AI binding or the REST inference API. The base Whisper model is described as multilingual speech recognition, speech translation and language identification. The turbo model page marks Batch: Yes and exposes task (transcribe | translate), language, vad_filter, initial_prompt, beam size and hallucination controls. ASR on Workers AI is rate-limited at 720 requests per minute. Free plans get 10,000 Neurons per day; usage above that on Workers Paid bills at $0.011 per 1,000 Neurons (unit prices still publish in dollars per audio minute). For live WebSocket STT through AI Gateway, Cloudflare documents Deepgram models (@cf/deepgram/nova-3, @cf/deepgram/flux) — a separate streaming SKU from Whisper inference.
Pricing converted to dollars per hour (Estimated)
| Path | Published rate | Implied $/hour of audio | Notes |
|---|---|---|---|
| LecSync Realtime Speech API | $0.006 / min all-in | $0.36 | STT + translation + speakers + digest included |
CF @cf/openai/whisper | $0.000453 / min (model page) | ~$0.0272 | Inference / file-style STT |
CF whisper-large-v3-turbo | $0.000513 / min (model page) | ~$0.0308 | Batch; optional EN translate task |
| CF pricing-table (both) | ~$0.0005 / min (rounded) | $0.03 | Audio table on Workers AI pricing |
Minute math (Estimated): whisper → 1000 hours ≈ $27.18; turbo → ≈ $30.78; rounded $0.0005 → $30. LecSync stays $0.006/min whether translation and speakers are on or off → 1000 hours = $360.
Side-by-side facts (checked October 26, 2026)
| LecSync Realtime Speech API | Cloudflare Workers AI Whisper | |
|---|---|---|
| Primary job | Live meeting/lecture captions + MT | Cheap edge/batch Whisper ASR |
| Price model | $0.006/min flat, all features | $0.000453–$0.000513/min (model pages) |
| Delivery | Managed WebSocket, preview→final | Workers binding / REST inference |
| Live streaming STT on CF | — | Whisper: no; Deepgram Nova-3/Flux via AI Gateway WS |
| Live text translation | Any two of 60, same socket | Whisper-style translate task (typically EN) on turbo; not live bilingual captions |
| Speakers on live stream | Every final segment | Not a documented Whisper feature |
| Live minutes / digest | digest: true | Not available |
| Languages | 60, auto-detected | Multilingual Whisper-family (docs) |
| Latency story | ~0.28 s first word; +~88 ms MT | Batch / request inference, not first-word socket latency |
| Rate limits (docs) | — | ASR 720 RPM |
| Free allocation | — | 10,000 Neurons/day; then $0.011/1k Neurons |
| Hosting | Managed SaaS, 7 gateway regions | Cloudflare edge / Workers AI |
What 1,000 hours cost (Estimated)
| Workload, 1,000 hours a month | LecSync | Cloudflare Workers AI Whisper |
|---|---|---|
| Same-language captions / transcripts | $360 | whisper ~$27; turbo ~$31; table ~$30 |
| Translated live captions (arbitrary pair) | $360 | Not on Whisper; DIY STT + MT (or separate CF streaming SKU) |
| Speakers with timestamps during the session | Included | Not documented on Whisper |
| Live minutes | Included | Not available |
| Post-call file transcription on the edge | Possible via your own stack | Whisper / turbo on Workers AI |
Cloudflare wins hard on raw Whisper dollars and Workers ecosystem fit. LecSync wins when the product is a live bilingual caption feed rather than a batch inference job. For a deeper flat-vs-add-on walkthrough, see what 1,000 hours cost once translation, speakers and summaries are added. For stitching your own STT + MT, see built-in translation vs a DIY pipeline. A nearby batch Whisper comparison is LecSync vs Groq Whisper.
When Cloudflare Workers AI Whisper is the better pick
- Batch or post-call Whisper transcription at roughly $0.00045–$0.00051 per audio minute.
- Workers / edge-native apps that already bind Workers AI and want ASR without a second vendor.
- Free Neuron headroom for light traffic (10,000 Neurons/day before Paid overage).
- Turbo batch controls —
task=translate,vad_filter, prompts and decoding thresholds on the large-v3-turbo model page. - High ASR request volume within the documented 720 RPM Automatic Speech Recognition limit.
When LecSync fits better
- Translated live captions on one socket — any pair of 60 languages, each translation tied to its segment.
- Speakers and recording-aligned timestamps during the live stream — meeting client with speakers and timestamps.
- Live minutes while the session runs (
digest: true) — live digest vs post-call summarization. - Predictable minute billing at $0.36 an hour with no separate translation SKU and no batch round-trip before captions appear.
- Classroom / study-abroad style feeds where captions must keep up as the lecturer switches languages, not after the recording uploads.
FAQ
Is Cloudflare Workers AI Whisper a realtime streaming caption API?
No. Model pages price Whisper as unit inference per audio minute; turbo is marked Batch. Cloudflare's AI Gateway Realtime WebSockets docs list Deepgram on Workers AI for live speech-to-text, not Whisper. Whisper is the cheap batch/file path; live streaming STT on CF is a different product line.
Why is LecSync more expensive per hour of audio?
LecSync's $0.36/hour includes live translation, speakers, glossary/polish and digest on the same connection. Cloudflare's ~$0.027–$0.031/hour buys Whisper transcription (plus an optional EN-oriented translate task on turbo). Building live bilingual captions on Whisper means adding your own streaming front end, MT, diarization and minutes — and owning that glue.
Does Workers AI Whisper return speaker labels?
The public Whisper model pages do not document live speaker diarization. For live speaker-tagged segments, LecSync exposes speakers: true on the realtime socket.
How do Neurons and the free allocation affect cost?
Workers AI bills in Neurons under the hood ($0.011 per 1,000 Neurons on Paid after 10,000 Neurons/day free). Model pages still publish dollars per audio minute, which is what the tables above use. Light daily usage may stay inside the free Neuron pool; production ASR usually hits Paid unit pricing.
Can I mix both?
Yes. Many teams run LecSync for the live classroom or meeting feed, then push archived audio to Workers AI Whisper for cheap edge re-transcription. If you need Cloudflare-native live streaming STT separately, that path is Deepgram on AI Gateway — not the Whisper models compared here.
Next step
Open the LecSync Realtime Speech API docs, create a connect session, and stream three seconds of silence to see preview and final segments on one socket — translation, speakers and digest included at $0.006 per audio minute.