AI_RetryError: Failed after 4 attempts. Last error: Provider returned error (Qwen3 Coder 480B A35B Instruct (free) through OpenRouter)

[OgnjenKop]

It appears that CLI is unusable with Qwen3 Coder 480B A35B Instruct free model, through OpenRouter. Few call in the workflow, it always results in the following error:

"AI_RetryError: Failed after 4 attempts. Last error: Provider returned error".

This is not the case with other free models.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Openrouter Qwen3 Coder giving errors #1459: Openrouter Qwen3 Coder giving errors - Similar error with Qwen3 Coder model through OpenRouter
• Qwen3 as well as Kimi K2 keep stopping mid-chat #1522: Qwen3 as well as Kimi K2 keep stopping mid-chat - Reports issues with Qwen3 model stopping and failing through OpenRouter
• Flow just stops when using openrouter models #1710: Flow just stops when using openrouter models - Reports OpenRouter models stopping after initial tool calls, specifically mentioning issues with Qwen3 Coder and Kimi K2
• OpenRouter AI_APICallError: Provider returned error name=UnknownError Server error #915: OpenRouter AI_APICallError: Provider returned error name=UnknownError Server error - Similar "Provider returned error" message with OpenRouter
• w is not an object #1045: w is not an object - Shows similar error patterns that might be related to provider issues

Feel free to ignore if none of these address your specific case.

[rekram1-node]

@OgnjenKop I think this is a provider issue, not sure what we can do about it on our end. I managed to talk to that model without issues here but I will say the free openrouter models are generally super heavily ratelimited and I wouldn't generally recommend them

[OgnjenKop]

@rekram1-node Interesting thing is that some other free, oss models are working with no issues. Deepseek R1T2 Chimera, for example.

Would it help if I try out qwen on other CLI tools to see how it performs there?

[rekram1-node]

@OgnjenKop what I was getting at is this error is due to the provider of the model, not us, they must have had some outage

[florath]

@OgnjenKop: you might also run into a known tool-calling problem with the (unpatched) qwen3-coder models:
https://docs.unsloth.ai/basics/qwen3-coder-how-to-run-locally

I experienced a couple of problems: #1809

[rekram1-node]

Closing since this was due to provider issue