ollama reasoning model won't give think progress

[NGC13009]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I believe this is a bug. I'll try to join the Continue Discord for questions
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows
- Continue version: 1.2.8
- IDE version: 1.104.2
- Model: qwen3:14b@ollama
- config:
  
models:
  - name: qwen3:4b&nothink - ollama
    provider: ollama
    model: qwen3:4b
    capabilities: [tool_use]
    roles:
      - chat
      - edit
      - autocomplete
      - apply
      - summarize
    requestOptions:
      extraBodyProperties:
        think: false
        keep_alive: -1
    defaultCompletionOptions:
      contextLength: 6144
      maxTokens: 2048

Description

Starting now, models with thinking capabilities can no longer provide thought processes. The thinking function for all models has been turned off and cannot be reactivated.

As can be seen from the debugging, the configuration sent to ollama by default disables the model's thinking functionality, even if the user has not set it this way:

....
 Options
{
  "contextLength": 16384,
  "maxTokens": 6144,
  "model": "qwen3:14b",
  "stop": [
    "<|im_start|>",
    "<|im_end|>"
  ],
  "reasoning": false
}

Normally, it should not be necessary to configure the option to disable the thinking process for each model; instead, this option should be left to the user's choice. However, I have not been able to find any related configurable options at the moment.

It might have been a code submission for this plugin that mistakenly set all models to disable thinking. We hope this issue can be fixed as soon as possible, and allow users to enable the thinking feature for models themselves (the thinking feature should be enabled by default for all models, unless the model is specifically designated as an autocomplete model or a model used for generating titles)

To reproduce

1. Use an ollama model with a thinking function
2. Have a conversation
3. Finding that the thinking process that should have been there has disappeared

Log output

CONTINUE CONSOLE：

...
 Options
{
  "contextLength": 16384,
  "maxTokens": 6144,
  "model": "qwen3:14b",
  "stop": [
    "<|im_start|>",
    "<|im_end|>"
  ],
  "reasoning": false
}
...

[bdougie]

Hi @NGC13009, thank you for the detailed bug report. I've investigated the codebase and can confirm this is indeed an issue with how reasoning support is detected for Ollama models.

Root Cause

The problem is in the modelSupportsReasoning function that determines whether to show the reasoning toggle (lightbulb icon) in the UI:
https://github.com/continuedev/continue/blob/main/core/llm/autodetect.ts#L139-L153

Currently, this function only returns true for:

1. Anthropic models (Claude)
2. DeepSeek-R models
3. Models with reasoning: true explicitly set in config

Ollama models are not detected as supporting reasoning, even though:

• Ollama correctly implements the think parameter: https://github.com/continuedev/continue/blob/main/core/llm/llms/Ollama.ts#L401
• Models like qwen3:14b have thinking capabilities built-in

Workaround

Until this is fixed, you can force enable reasoning for your Ollama model by adding reasoning: true to your completion options:

models:
  - name: qwen3:14b - ollama
    provider: ollama
    model: qwen3:14b
    completionOptions:
      reasoning: true  # This forces the reasoning toggle to appear
      contextLength: 16384
      maxTokens: 6144

This will make the reasoning toggle (lightbulb icon) appear in the chat input toolbar, allowing you to enable/disable thinking for your model.

Additional Context

The reasoning: false you're seeing in debug logs for edit/apply operations is intentional - reasoning is disabled for code edits to prevent thinking output during modifications:

• https://github.com/continuedev/continue/blob/main/core/edit/recursiveStream.ts#L46
• https://github.com/continuedev/continue/blob/main/core/edit/recursiveStream.ts#L83

For regular chat conversations, the reasoning flag should respect the toggle state when available.

Proposed Fix

The modelSupportsReasoning function needs to be updated to detect Ollama models with thinking capabilities. This could be done by:

1. Adding detection for known thinking-capable models (qwen3, qwen2.5, deepseek-r1, etc.)
2. Or checking if provider is 'ollama' and the model name contains certain patterns
3. Or allowing Ollama to report thinking capabilities through its API

Would you mind testing the workaround and confirming if it enables the thinking functionality for you?