Code Interpreter Rate Limits

Are there rate limits for Code Interpreter? They seem to be different depending on what model I’m using.

gpt-3.5-turbo-1106 significantly less so than gpt-4-1106-preview.

I can’t see any documentation on it. Can you?

The only rate limits you are given is by tokens and model call, per AI model.

Code interpreter uses a minimum of two AI model calls:

  • one for it to write the code the sandbox will execute,
  • one for writing the response to you about the results.

Then it may be running multiple calls outside of your control, trying to fix multiple traceback errors or get multi-step results. This will amplify the usage statistics, along with the continued reuse of conversational context tokens.

1 Like

Hmm it can’t be that then.

In the last couple of days it’s either failed to trigger code_interpreter in obvious situations (or even when prompted) or I’ve received an ‘internal limit’ error.

Tends to happen in the playground when I run an assistant consecutively in a short period of time, like a rate limit (but super super low).

“Internal limit” as an assistant status may be a report of those iterations being run out of control: the AI calling python with the same thing over and over. And since you can’t see, you pay, but can’t debug.

Step 1: don’t use the still-broken 1106 preview model.

1 Like