qwen3.5:9b sometimes prints out tool call instead of executing it

[Kangaroux]

workaround: use ollama 0.17.5

What is the issue?

Using the model from ollama.com and Opencode 1.2.24. This happens fairly often and halts the work the agent was doing.

Thinking: I need to mock Valkey for tests. Let me check how the app initializes Valkey and update TESTING.md with this information.
<tool_call>
<function=bash>
<parameter=command>
grep "redacted" | head -20
</parameter>
<parameter=description>
Find Valkey initialization
</parameter>
</function>
</tool_call>

$ ollama ls
NAME          ID              SIZE      MODIFIED     
qwen3.5:9b    6488c96fa5fa    6.6 GB    25 hours ago

$ ollama show qwen3.5:9b
  Model
    architecture        qwen35    
    parameters          9.7B      
    context length      262144    
    embedding length    4096      
    quantization        Q4_K_M    
    requires            0.17.1    

  Capabilities
    completion    
    vision        
    tools         
    thinking      

  Parameters
    presence_penalty    1.5     
    temperature         1       
    top_k               20      
    top_p               0.95    

  License
    Apache License               
    Version 2.0, January 2004    
    ...

$ ollama show --modelfile qwen3.5:9b
# Modelfile generated by "ollama show"
# To build a new Modelfile based on this, replace FROM with:
# FROM qwen3.5:9b

FROM /usr/share/ollama/.ollama/models/blobs/sha256-dec52a44569a2a25341c4e4d3fee25846eed4f6f0b936278e3a3c900bb99d37c
TEMPLATE {{ .Prompt }}
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER temperature 1
PARAMETER top_k 20
PARAMETER top_p 0.95
PARAMETER presence_penalty 1.5

Relevant log output

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.17.7

[Kangaroux]

Possibly related?

• Qwen 3.5 27B: Tool calling completely non-functional and repetition penalties silently ignored #14493
• Qwen3 tool calling via /api/chat tools parameter: malformed tool definitions (and soft switch thinking issue) #14601
• model/renderers: fix Qwen 3.5 tool call format and unclosed think tags #14603

[nandakho]

yeah, i experienced this issue too, while running opencode with ollama 0.17.6 and 0.17.7.
seems related with how thinking tag open/close.
however, using 0.17.5 i haven't encountered this issue, so i'm sticking with it until further update

[TheSpaceGod]

Having same issue as well. Can confirm switching to 17.5 also fixed the issue for me

[Kangaroux]

yeah, i experienced this issue too with opencode on 0.17.6 and 0.17.7. seems related with how thinking tag open/close. however, using 0.17.5 i haven't encountered this issue, so i'm sticking with it until further update

You meant ollama yeah?

[TheSpaceGod]

yeah, i experienced this issue too with opencode on 0.17.6 and 0.17.7. seems related with how thinking tag open/close. however, using 0.17.5 i haven't encountered this issue, so i'm sticking with it until further update

You meant ollama yeah?

Yes, ollama

[nandakho]

yeah, i experienced this issue too with opencode on 0.17.6 and 0.17.7. seems related with how thinking tag open/close. however, using 0.17.5 i haven't encountered this issue, so i'm sticking with it until further update

You meant ollama yeah?

Yes, ollama. Edited to avoid confusion.

[aldi7th]

Downgrade to 0.17.5 works for me, and I tried 0.17.8-rc3, still encountered this issue.

[metal3d]

I don't know what changed in 0.17.6 but I have the problem too and I needed to downgrade to 0.17.5 to have the model using tools correctly.

[mnlipp]

The problem still exists with 0.18.

[TheSpaceGod]

Related PR that likely introduced the bug which other tickets are linking to: #14605

[LR1093141031]

0.18.2 still

[krvkir]

I get an error which looks similar but is probably of different origin. I try to use ollama in jupyter via jupyter-ai. Tools work fine with gemini3.1 hosted by Google, but when I use gemini3 over ollama (running in the cloud), I get this:

{"name": "get_active_notebook", "arguments": {"username": "<my user id>"}}

and the model halts after this message.

The same happens with ollama with other large models in the cloud and with small models running locally.

Ollama version is 0.18. I tried downgrading it to 0.17.5, but the problem was still there.

Interesting is that the same models happily use tools in emacs+gptel, so that's possibly a jupyter-ai bug.