Roo-Code: API Request Failed - rate limit exceeded

Which API Provider are you using?

VS Code LM API

Which Model are you using?

Claude 3.5 Sonnet

What happened?

I keep getting this rate limit exceeded error. If I wait for sometime ~5 mins, I am able to start the command again Claude 3.5 but I thought that there was not limits on this that I could keep using it without hitting any limits.

Steps to reproduce

1.Run the model for about an hour 2. 3.

Relevant API REQUEST output


Additional context

No response

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 25

Most upvoted comments

I received a response to my ticket from GitHub customer service. They told me that repeated rate limit exceeded messages will result in a ban on personal GitHub account. This is stated in their terms of service. Let’s say the $10/month subscription is only for GitHub Copilot Pro and not for anything else, like Roo-Code.

I have the same issue: API Request failed: rate limit exceeded

Does it come from Github Copilot? I’m using a Pro version, so I should have unlimited messages.

I think we are encountering this because the copilot subscription wasn’t supposed to be used this way. The unlimited messages are only for their chat interface, not the API requests outside of it, that’s what explains the hidden rate limit in my opinion, and I am assuming they will patch/update their usage policies because literally it would have costed me $50 per day if I was using Anthropic/Openrouter instead of the $10 of copilot per month for my Sonnet usage.

Tried using the new exponential fall back feature last night and really it’s not needed because the rate limit is based on token usage. I used to be able to use about 5M tokens before they cut me off and now it’s down to 2M and then they shut me off. Really sucks that MSFT is doing this since my company pays a lot of money for all of our devs to have CoPilot accounts

How many tokens per day do you use on average?

Depending on the work of course but on average maybe 10M. Now Github/MS seem to have caught that their API is being used not in the way they were intended thus reduced the limits drastically (used to get about 5M tokens per hour now they don’t exceed 1.5M)

I think we can report this to GitHub Support, as there doesn’t seem to be dedicated support for Copilot. You can find more details here: Contacting GitHub Support

Here’s the direct link to create a ticket: GitHub Support Ticket

I’ve already created a ticket and will let you know if I receive a response.

I have the same issue and I have a Pro subscription. How can I report it to GitHub Copilot support?

I understand it probably comes from Github Copilot, status = 429. Would be nice to know if there is any additional information, like the time when the limit will go away?

☝ will go out in the next release

My suggestion is to replace linear retry with exponential back off implementation.

That makes a lot of sense!

What exactly does that mean?

I just opened a PR for this. It means that if you have consecutive failures, we will wait longer each time. In what I’m proposing the first retry would be after 5 seconds, the second after 10 more seconds, the third after 20s, the fourth after 40s, etc.

I think we are encountering this because the copilot subscription wasn’t supposed to be used this way. The unlimited messages are only for their chat interface, not the API requests outside of it, that’s what explains the hidden rate limit in my opinion, and I am assuming they will patch/update their usage policies because literally it would have costed me $50 per day if I was using Anthropic/Openrouter instead of the $10 of copilot per month for my Sonnet usage.

I tested the Copilot chat interface and I also got a rate limit message: “Sorry, your request was rate limited. Please wait 4 seconds before trying again”.

Github copolit is not unlimited anymore. Below could be relevant issues.

Image

Could you inform us on where we can report this to CoPilot? I am on the Pro plan as well and still facing this issue.