Aider shows no response when OpenRouter free models run out of rate limit

[Astlaan]

Aider version: 0.74.2
Python version: 3.12.7
Platform: Windows-11-10.0.26100-SP0
Python implementation: CPython
Virtual environment: Yes
OS: Windows 11 (64bit)
Git version: git version 2.44.0.windows.1

Aider v0.74.2
Main model: openrouter/openai/o3-mini with diff edit format
Weak model: openrouter/openai/gpt-4o-mini
Git repo: .git with 8 files
Repo-map: using 4096 tokens, auto refresh

PS C:\Users\user\Documents\Code\windows-openai-whisper> aider --model openrouter/openai/o3-mini
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
Aider v0.74.2
Main model: openrouter/openai/o3-mini with diff edit format
Weak model: openrouter/openai/gpt-4o-mini
Git repo: .git with 8 files
Repo-map: using 4096 tokens, auto refresh
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
> are you there?



Tokens: 2.9k sent, 0 received. Cost: $0.0032 message, $0.0032 session.
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
>

[VELIKII-DIVAN]

Aider v0.74.2
Main model: deepseek/deepseek-reasoner with architect edit format, prompt cache, infinite output
Editor model: claude-3-5-sonnet-20241022 with editor-diff edit format
Weak model: deepseek/deepseek-chat
Git repo: .git with 88 files
Repo-map: using 4096 tokens, auto refresh

Python version: 3.12.7
Platform: Windows 10 Pro 22H2 19045.5371
OS: Windows 10 (64bit)
Git version: git version 2.20.0.windows.1

The same problem and in the architect mode too

[paul-gauthier]

Thanks for trying aider and filing this issue.

Does it help to use --no-pretty or --verbose or both?

[VELIKII-DIVAN]

aider_huge_logs.txt here is full log. I removed only api keys.
I think this bug happens after /add command

[paul-gauthier]

I can not reproduce this problem.

The lack of response from deepseek/deepseek-reasoner is likely due to the main deepseek API being flakey/down/overloaded.

[twinity1]

I got this behaviour when I had no credits remaining on my OpenRouter account.

OpenRouter returns something like this when there are no credits remaining:

{
"error": {
"message": "More credits are required to run this request. 64000 token capacity required, 1644 available. To increase, visit https://openrouter.ai/credits and add more credits",
"code": 402
}
}

However, Aider doesn't display any error message, which might be confusing.
I think the deducted credits are calculated from the model's metadata that is packed inside Aider. But in reality, it doesn't deduct anything.

[diamond-lizard]

When this happened to me in Aider no matter which free model I chose, I ran the following test API call (using the demo code from the Quickstart section of the OpenRouter docs:

 OPENROUTER_API_KEY=my-seekrit-key ; curl https://openrouter.ai/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -d '{
  "model": "microsoft/phi-3-medium-128k-instruct:free",
  "messages": [
    {
      "role": "user",
      "content": "What is the meaning of life?"
    }
  ]
}'

and received this reply:

{"error":{"message":"Rate limit exceeded: free-models-per-day","code":429,"metadata":{"headers":{"X-RateLimit-Limit":"200","X-RateLimit-Remaining":"0","X-RateLimit-Reset":"1741305600000"}}}}

Adding credits did not help, because (I guess) the free model rate limit applies regardless of whether you have credits or not.

It would be nice if Aider recognized that such errors resulted from either lack of credits or rate limiting and alerted the user accordingly instead of merely saying that 0 tokens were received.

[paul-gauthier]

Thanks for the good diagnostic info. I can reproduce this as a litellm bug. See issue above.