DeepSeek Thinking Mode: Enable, Read & Compare

1. User sends prompt with thinking: { type: "enabled" }
2. Model generates reasoning_content (visible CoT tokens)
3. Model generates content (final answer)
4. Both are returned in the response
5. reasoning_content is billed as output tokens
6. In multi-turn: reasoning_content may or may not need to be passed back (depends on context)

from openai import OpenAI

client = OpenAI(
    api_key="<DeepSeek API Key>",
    base_url="https://api.deepseek.com"
)

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Solve this complex problem..."}],
    reasoning_effort="high",  # or "max"
    extra_body={"thinking": {"type": "enabled"}}
)

# Access reasoning content
reasoning = response.choices[0].message.reasoning_content
answer = response.choices[0].message.content

print(f"REASONING:\n{reasoning}\n")
print(f"ANSWER:\n{answer}")

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.deepseek.com/anthropic",
    api_key="<DeepSeek API Key>"
)

response = client.messages.create(
    model="deepseek-v4-pro",
    max_tokens=4096,
    system="You are a helpful assistant.",
    messages=[{"role": "user", "content": "Solve this complex problem..."}],
    output_config={"effort": "high"}  # or "max"
)

if answer_is_wrong:
    print(f"MODEL'S REASONING:\n{reasoning}")
    # Look for: faulty assumptions, skipped steps, premature conclusions
    # The error will be visible in the reasoning chain

# Long, structured reasoning = high confidence
# Short, hand-wavy reasoning = low confidence

if len(reasoning) < 200 or "clearly" in reasoning:
    print("Warning: Model may be overconfident or reasoning insufficiently")

User prompt: "Verify your own reasoning before giving the final answer."

This prompts DeepSeek to include a self-check in reasoning_content:
"Let me verify: assumption A is correct because... assumption B holds given...
Wait — assumption B conflicts with constraint C. Let me reconsider..."

stream = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=messages,
    stream=True,
    reasoning_effort="high",
    extra_body={"thinking": {"type": "enabled"}}
)

reasoning_buffer = ""
answer_buffer = ""

for chunk in stream:
    delta = chunk.choices[0].delta
    if delta.reasoning_content:
        reasoning_buffer += delta.reasoning_content
    elif delta.content:
        answer_buffer += delta.content