What happened?
Intermittently, the model stream returns an abnormal response (HTTP 200 but effectively empty), and Qwen Code fails with:
API Error: Model stream ended without a finish reason.
Logs:
Observations:
hasFinishReason=falsecontentTextLength=0hasUsageMetadata=trueresponse_id=""- very small
output_token_count (e.g. 3)
A related log may also appear:
[CHAT_RECORDING] Error saving assistant turn: Error: partOrString cannot be an empty array
This error interrupts ongoing automated execution in Qwen Code.
What did you expect to happen?
When this transient empty-stream condition happens, Qwen Code should automatically retry and recover whenever possible, instead of failing. From practical usage, retrying shortly after often succeeds.
Client information
- Qwen Code:
0.11.0 (a31cf757)
- Runtime:
Node.js v22.22.0 / npm 10.9.4
- IDE client:
VS Code
- Model:
bailian/glm-5
- Session ID:
a61da563-3974-478a-9518-2ec6cd2b9c5f
- Proxy:
no proxy
- Memory:
261.2 MB
Anything else we need to know?
The issue is intermittent, but once it happens it can break long-running automated analysis/tool-use flows. Manual retry after a short wait often succeeds, which suggests a transient upstream stream anomaly rather than a deterministic prompt/content failure.
What happened?
Intermittently, the model stream returns an abnormal response (HTTP 200 but effectively empty), and Qwen Code fails with:
API Error: Model stream ended without a finish reason.Logs:
Observations:
hasFinishReason=falsecontentTextLength=0hasUsageMetadata=trueresponse_id=""output_token_count(e.g.3)A related log may also appear:
[CHAT_RECORDING] Error saving assistant turn: Error: partOrString cannot be an empty arrayThis error interrupts ongoing automated execution in Qwen Code.
What did you expect to happen?
When this transient empty-stream condition happens, Qwen Code should automatically retry and recover whenever possible, instead of failing. From practical usage, retrying shortly after often succeeds.
Client information
0.11.0 (a31cf757)Node.js v22.22.0 / npm 10.9.4VS Codebailian/glm-5a61da563-3974-478a-9518-2ec6cd2b9c5fno proxy261.2 MBAnything else we need to know?
The issue is intermittent, but once it happens it can break long-running automated analysis/tool-use flows. Manual retry after a short wait often succeeds, which suggests a transient upstream stream anomaly rather than a deterministic prompt/content failure.