Whisper JAX vs WhisperAI
The Whisper JAX demo on Hugging Face says Running, but as of November 13, 2026, it times out with 504 Gateway Time-out instead of loading. If you came for a fast transcript from a web page, WhisperAI does that from a browser tab with nothing to install, and an hour of audio usually comes back in under a minute.
By Eman Khoubian, Founder & CEO
Is Whisper JAX online on Hugging Face?
Not in a way you can use. We tried it three times in five minutes, and each time it answered with a 504 Gateway Time-out.
You're not the first to see it. In July 2025, one user wrote: "I tried to access whisper-jax from different networks and different countries ... Still can't access." In February 2026, another replied: "still same issue : 504 Gateway Time-out."
The Space was last changed in September 2024, and the code on GitHub in April 2024.
Transcribe what you'd have given Whisper JAX
In WhisperAI, all of it happens in a browser tab. There's no model to pick, no GPU to rent and no JAX version to match.
| Job | WhisperAI | Whisper JAX demo, when it ran |
|---|---|---|
| Transcribe a file | Upload audio or video up to 10 hours or 5 GB | Files up to 1,000 MB, on Whisper large-v3 |
| Transcribe a YouTube link | Import from YouTube turns the video's captions into a transcript, for videos that have them. It uses none of your minutes. | Downloaded the video's audio and transcribed it, for videos up to 2 hours |
| Translate | Transcribe in 100+ languages, then translate the transcript into any of 87 | Into English |
| Timestamps | A time on every sentence, plus SRT and VTT subtitle files | Optional times on each chunk of text |
| Speaker labels | One switch at upload, on every plan | None |
How fast is Whisper JAX?
Fast, on the right hardware. Its README says it "runs over 70x faster" than OpenAI's own code. That figure comes from a Google TPU v4-8. On the same NVIDIA A100 graphics card OpenAI's code ran on, Whisper JAX took 75.3 seconds for an hour of audio against 1001.0, which is about 13x.
And you have to rent a TPU from Google Cloud. A Hacker News commenter called that the main drawback in 2023: "the performance comes from running on Google TPUs, which are much more expensive than GPUs." The same commenter pointed to faster-whisper for ordinary GPUs and whisper.cpp for CPUs and Apple Silicon. Our comparison with OpenAI Whisper covers running Whisper on your own machine.
Running Whisper JAX yourself today takes some work. The README says it was tested on Python 3.9 and JAX 0.4.5, and its setup file needs a Hugging Face Transformers version below 4.35. Newer JAX versions crash with "module 'jax.core' has no attribute 'NamedShape'", an issue open since January 2025, with two fixes still unmerged. JAX itself lists no NVIDIA GPU support on native Windows and calls Apple's Mac GPUs experimental. If you'd rather not set any of that up, here's how to use Whisper without coding.
Questions people also ask
Is Whisper JAX free?
Yes. The code is free under the Apache 2.0 license. The hardware it needs, a GPU or a rented TPU, is on you.
Is Whisper JAX made by OpenAI?
No. OpenAI made the Whisper model. Whisper JAX is Sanchit Gandhi's port of it to JAX, built on Hugging Face's Transformers code. WhisperAI isn't part of either: we're an independent company, and AssemblyAI transcribes the files you upload.
What is the difference between Whisper JAX and WhisperX?
Whisper JAX is about speed: it runs Whisper on JAX, fastest on a TPU. WhisperX adds a time for each word and speaker labels, using other models alongside Whisper. WhisperX vs WhisperAI has more.
Is WhisperAI free?
There's a free plan with 5 minutes a month and no card. Premium is $14.99 a month for 120 minutes, and Business Pro is $24.99 a month with no minute limit.
Transcribe your file now, with nothing to install
Start on the free plan, no card. Upgrade when you need more minutes.
Try WhisperAI free