Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading from 5.7 to above, like 5.20, openclaw reports LLM request timed out when answer a long output question by ollama + qwen 3.5:27b. But same question is answered well directly in ollama + wen 3.5:27b. I even use default parameters under model section in openclaw.json, but still not help.

"models": {
"ollama/qwen3.5:27b": {
"params": {
"cacheRetention": "long",
"offload_kv": true,
"stream": true,
"use_mmap": true
}
},

....

"timeoutSeconds": 1800,
"thinkingDefault": "medium",

...

"models": {
"mode": "merge",
"providers": {
"ollama": {
"baseUrl": "http://xxx.xxx.xxx.xxx:11434",
"api": "ollama",
"apiKey": "OLLAMA_API_KEY",
"timeoutSeconds": 1800,
...
"models": [
{
"id": "qwen3.5:27b",
"name": "qwen3.5:27b",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 262144,
"maxTokens": 8192,
"params": {
"num_ctx": 262144
}
}
]
},

Steps to reproduce

1. when return to old version: 5.7 or 5.8, it's ok.
2. only happened to upgrade to 5.19, or 5.20.

Expected behavior

Should work without LLM request timed out for ask a complex question with long output. First it's workable directly drive by ollama + qwen 3.5:27b local model. Second it's workable in old openclaw version. Third even remove existing parameters inside openclaw.json about this model's settings, such as: "num_ctx", "num_predict", etc, still not report LLM request timed out. So that's not model itself issue and it's also not parameters settings issue. Last it's ok when ask simple questions. So guess it's a timeout settings. But we have timeoutSeconds settings in openclase.json with 1800. But seems our case report timeout when over 6 minutes.

Actual behavior

refer to above.

OpenClaw version

5.20

Operating system

centos stream 8 with latest version. No version and os packages change.

Install method

No response

Model

ollama + local model: qwen 3.5:27b

Provider / routing chain

openclaw -> qqbot channel

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response