Thinking

Thinking-capable models emit a thinking field separating reasoning trace from final answer.

Use this to: audit model steps, animate "thinking" in UI, or hide the trace when you only need the final response.

Supported Models

Note: Thinking is enabled by default in CLI and API for supported models.

API Usage

Non-Streaming

curl http://localhost:11434/api/chat -d '{
  "model": "deepseek-r1:1.5b",
  "messages": [{"role": "user", "content": "How many r letters in strawberry?"}],
  "think": true,
  "stream": false
}'

Response:

{
  "message": {
    "role": "assistant",
    "thinking": "Let me count: s-t-r-a-w-b-e-r-r-y. The letter r appears at positions 3, 8, 9. So 3 times.",
    "content": "There are 3 letter r's in 'strawberry'."
  }
}

GPT-OSS (Level-Based)

curl http://localhost:11434/api/chat -d '{
  "model": "gpt-oss",
  "messages": [{"role": "user", "content": "What is 1+1?"}],
  "think": "low"
}'

Note: GPT-OSS ignores true/false — must use "low", "medium", or "high".

Python SDK

from ollama import chat

Non-streaming
response = chat(
    model='deepseek-r1:1.5b',
    messages=[{'role': 'user', 'content': 'How many r letters in strawberry?'}],
    think=True,
    stream=False,
)
print('Thinking:\n', response.message.thinking)
print('Answer:\n', response.message.content)

Streaming with thinking
stream = chat(
    model='qwen3',
    messages=[{'role': 'user', 'content': 'What is 17 x 23?'}],
    think=True,
    stream=True,
)

in_thinking = False
for chunk in stream:
    if chunk.message.thinking and not in_thinking:
        in_thinking = True
        print('Thinking:\n', end='')
    if chunk.message.thinking:
        print(chunk.message.thinking, end='')
    elif chunk.message.content:
        if in_thinking:
            print('\n\nAnswer:\n', end='')
            in_thinking = False
        print(chunk.message.content, end='')

JavaScript SDK

import ollama from 'ollama'

// Non-streaming
const response = await ollama.chat({
    model: 'deepseek-r1',
    messages: [{ role: 'user', content: 'How many r letters in strawberry?' }],
    think: true,
    stream: false,
})
console.log('Thinking:\n', response.message.thinking)
console.log('Answer:\n', response.message.content)

CLI Quick Reference

# Enable thinking for a single run
ollama run deepseek-r1 --think "Where should I visit in Lisbon?"

Disable thinking
ollama run deepseek-r1 --think=false "Summarize this article"

Hide trace while still using thinking model
ollama run deepseek-r1 --hidethinking "Is 9.9 bigger or 9.11?"

Interactive session toggle
/set think
/set nothink

GPT-OSS levels
ollama run gpt-oss --think=low "Draft a headline"
ollama run gpt-oss --think=medium "Analyze this code"
ollama run gpt-oss --think=high "Prove this theorem"

mindX Integration

The existing OllamaAPI.generate_text() already supports think via kwargs:

# Already works via api/ollama/ollama_url.py line:
for key in ["format", "system", "template", "raw", "suffix", "images", "think"]:
    if key in kwargs:
        payload[key] = kwargs[key]
result = await ollama_api.generate_text(
    prompt="Analyze the autonomous improvement cycle for inefficiencies",
    model="deepseek-r1:1.5b",
    think=True  # Passes through to payload
)
The response includes thinking trace in the response text
For chat endpoint, thinking is in message.thinking field

Thinking for Agent Tasks

Different agents benefit from different thinking levels: