Thinking

The settings

Any other value is refused with 400 invalid_reasoning_effort, naming the ones the model takes.

Turn thinking on

Ask for thinking on hard questions: planning, maths, code, long instructions. Give the reply room, since thinking counts toward max_tokens.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://nokeep.ai/v1",
    api_key=os.environ["NOKEEP_API_KEY"],
)

reply = client.chat.completions.create(
    model="deepseek-v4.1-flash-uncensored",
    reasoning_effort="high",
    max_tokens=4000,
    messages=[{"role": "user", "content": "Plan a week of meals for four on $80."}],
)
print(reply.choices[0].message.content)

With thinking on, set max_tokens to 1,000 or more. Under that, the model can spend every token thinking and send back an empty reply, which is still billed.

Read the thinking

The thinking comes back apart from the answer: in message.reasoning_content on a whole reply, and in delta.reasoning_content while streaming, before the answer’s first word. Show it, log it on your side, or ignore it.

usage.completion_tokens_details.reasoning_tokens says how many of the output tokens were thinking.

In apps

Fiction apps such as SillyTavern and JanitorAI send a short reply limit and no thinking setting, so DeepSeek gives them a fast, whole reply. Apps that let you add body fields, such as SillyTavern, can send reasoning_effort to turn thinking on; give the reply 1,000 tokens or more when you do.