Workers AI Speech Recognition request limits

Developers Workers AI
1

Hi there. We are happy with all the Cloudflare services that we used.

Recently, we started with Workers AI Speech Recognition (Whisper) API enpoint and observerd the following:

  • MP3 file // 128kbps // 2-channels // 00:00:45 (45 seg) lenght // 705kb size: HTTP request ok (word_count: 95).
  • MP3 file // 128kbps // 2-channels // 00:04:28 (4 mins 28 seg) lenght // 4200 kb size: HTTP error

{"errors":[{"message":"AiError: AiError: Invalid or incomplete input for the model: failed to decode JSON: Request is too large","code":3010}],"success":false,"result":{},"messages":[]}

We are wondered to know what are the limit request to this API endpoint?

https://api.cloudflare.com/client/v4/accounts/ACOUNT_REDACTED/ai/run/@cf/openai/whisper

Thanks in advanced. Best regards.

2 Likes
2

So stoked you’ve had a good experience with Cloudflare so far!

Right now, we’re constantly updating (and improving!) some of our limits. Since they’re changing frequently they are not something we publish. In the coming months, as these limits become more consistent we’ll start sharing them in our docs.

1 Like
3

Hi @rrobinett , thanks for the feedback. We appreciate this.

We will visit the documentation regularly for further updates.

Regards.

4

Hello, @rrobinett , what’s the longest time so far? Right now Cloudflare Worker can’t use the OpenAI Whisper API, so we’re looking for other solutions. I look forward to hearing from you, thank you.

5

Hey @rrobinett do you know if there is an update on the worker limit?
Looks like the new AI unit pricing is set to start this coming Nov yet the current file size limit on the processing worker is a deal breaker for most applications.
Also, maybe you can recommend on a workaround until this is resolved?
Thanks!

6

Looking for a solution for this as well. trying a 25MB 10min mp3 audio also fails with a “network dropped” error for me:

{"errors":[{"message":"Error: Network connection lost.","code":6001}],"success":false,"result":{},"messages":[]}
7

When running it from within a worker and invoking the model directly via env.AI.run we get a weird internal error stack trace:

AiError: undefined: undefined
 at Ai._parseError (cloudflare-internal:ai-api:76:20)
 at async Ai.run (cloudflare-internal:ai-api:57:23)
 at async Object.fetch (index.js:24:24)
8

Even though GROQ mentions a limit of 25MB for a file (not sure if it’s a model limit or just an infra limit on their end) it processes my 25.7MB file just fine:

image
image1813Ă—250 37.5 KB

I subscribed to the Cloudflare Workers Paid plan to see if that would make any difference but no deal.

1 Like
9

As of April 2025, this issue still persists - tested out with 37 min 36 MB file and getting the same unhelpful error message.

It’s been a year, wonder if @rrobinett has any updates on the limits? I’m unable to find any information in the docs regarding the limits.

10

Hi @zsheng , @roberto12 .

Please, keep aware that the OpenAI Whisper model has a 25Mb file size limitation. That’s why other providers also inherit the same limitation.

As workarround, you can try convert the audio file to another format. I have see other Internet users claim 40Mb MP3 file compress to 8Mb in M4A format. Or try use Opus codec + OGG container.

1 Like