Opencode + Qwen gives JSON output instead of writing to disk

[dvictori]

Description

I'm using OpenCode 1.15.11 with Qwen2.5-code:7b running with ollama on my desktop.

I asked for a simple python script, saved into a file. Instead of creating the file on disk, OpenCode returned the JSON for the write command.

I'm using the build agent and here is my ./.config/opencode/opencode.jsonc. I understand that I have given all the necessary permissions to the model

{
    "$schema": "https://opencode.ai/config.json",
    "permission": {
	"edit": "ask",
	"bash": "ask",
	"webfetch": "allow",
	"read": "ask",
	"write": "ask"
    },
   "provider": {
       "ollama": {
	   "name": "ollama",
	   "npm": "@ai-sdk/openai-compatible",
           "options": {
               "baseURL": "http://localhost:11434/v1"
           },
           "models": {
      	       "qwen2.5:7b": { "name": "Qwen2.5 7b" },
      	       "qwen2.5-coder:7b": { "name": "Qwen2.5 Code 7b" },
	       "qwen3.5:9b": { "name": "Qwen3.5 9b" }
           }
       }
   }
}

Plugins

No response

OpenCode version

1.15.11

Steps to reproduce

1. Install ollama and Qwen2.5-code:7b (using 16k context size)
2. Instruct opencode/Qwen2.5 to create the file

Screenshot and/or share link

No response

Operating System

Ubuntu

Terminal

No response

[github-actions]

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

• Ollama (qwen2.5-coder): tool calls (edit/write) show as executed but no files are created/modified #7030: Ollama (qwen2.5-coder): tool calls (edit/write) show as executed but no files are created/modified — same model, same symptom (JSON tool payload printed but files not written to disk)
• local model keeps returning json and doesn't create any files #18454: local model keeps returning json and doesn't create any files — same root symptom with Ollama + Qwen
• not compatible to ollama/qwen2.5-coder:7b #10402: not compatible to ollama/qwen2.5-coder:7b — same model/provider combination causing tool call issues

[QQi44]

I have this issue too. It happens with windows

[QQi44]

u can sove this issue by Increase context length

[Virgula0]

I'm using a bigger context length and it doesn't work on each of the following models, context size: 32k. Same problem of the OP. Version 1.15.10

{
  "$schema": "https://opencode.ai/config.json",
  "model": "ollama/qwen2.5-coder:7b-32k",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (Local Tool-Capable)",
      "options": {
        "baseURL": "http://127.0.0.1:11434/v1"
      },
      "models": {
        "qwen2.5-coder:7b-32k": {
          "name": "Qwen 2.5 Coder 7B (32k Context)"
        },
        "qwen2.5-coder:7b": {
          "name": "Qwen 2.5 Coder 7B (Default Context)"
        },
        "llama3.1:8b-32k": {
          "name": "Llama 3.1 8B (32k Context)"
        },
        "llama3.1:8b": {
          "name": "Llama 3.1 8B (Default Context)"
        }
      }
    }
  }
}

Update:

adding

"permission": {
	"edit": "ask",
	"bash": "ask",
	"webfetch": "allow",
	"read": "ask",
	"write": "ask"
    },

that was missing, llama3.1:8b-32k and lama3.1:8b seem to work. You need to specify the full path location where to write files but at least seems to produce something. Qwen still completely broken

With working llama 3.1 what i'm noticing is that it is not able to write in correct locations:

sh-5.3$ pwd
/tmp/we
sh-5.3$ cat ../test.txt 
test

[dvictori]

I tested with both default and 16k context length and none worked

[Virgula0]

I tested with both default and 16k context length and none worked

I found something working:

google/gemma-4-E4B

But with continue.dev visual studio code extension, not using opencode. Seems to have a lot of problems with small models