qwen-code finishes session abruptly after trying to call a tool

[nae104150]

What happened?

A session with any model finishes abruptly after trying to call a tool. I tried several models from different providers like qwen/qwen3-coder-plus from Alibaba, qwen/qwen3-coder-next and from DeepInfra, and finally moonshotai/kimi-k2.5. The behaviour is same. Please, take a look at the screenshot

What did you expect to happen?

The session doesn't finish abruptly and the agent continues to call tools.

Client information

Qwen Code: 0.12.6 (ac30c98)
Runtime: Node.js v22.18.0 / npm 10.9.3
OS: linux x64 (6.17.0-14-generic)
Auth: API Key - openai
Base URL: https://openrouter.ai/api/v1
Model: moonshotai/kimi-k2.5
Session ID: 266edd6d-2fdd-400f-af10-668b9f028f74
Sandbox: no sandbox
Proxy: no proxy
Memory Usage: 254.2 MB

Login information

No response

Anything else we need to know?

The behaviour is the same on Ubuntu and MacOS.
A couple of days earlier the agent worked without any issue.

[onwsk8r]

I've had issues with this as well, starting around 0.12.3. I have multiple model providers configured, and tool calling works fine when using qwen3.5-plus, served by Alibaba out of Singapore, but when I switch to OpenRouter (both using OpenAI auth- I'm using my own Alibaba account, not a qwen.ai coding account/etc), I get the result above.

Asking the model to diagnose results in a confused model: it seems to be able to see more than is shown in the screenshot (ie it knows there is a failed tool call), but cannot understand why the call failed. It just wants to try again.

[nae104150]

@tanzhenxin, @simon100500 Thank you for a quick response!

I reinstalled the npm package and tried again. Yet nothing have changed.
Should we wait for the version update?

[tanzhenxin]

@nae104150 @onwsk8r Sorry for the trouble, we did fixed the tool use issue when using OpenRouter provider, and the fix has been merged into main.
However, due to our release workflow, the next release would be a major feature release with a lot cool features, and we would test it internally a little longer than usual. The next release is expected in Sunday or next Monday.
If you are really in need of it, you can try our preview version (starting with prefix 0.13.0-preview), it should be ready in a few hours.