DeepSeek Reasoning Conversation Format (Chat Completions-style)

Deepseek Reasoning Conversation Format (Chat Completions Style)

Overview

Official Documentation

Reasoning model (deepseek-reasoner)

📝 Introduction

Deepseek-reasoner is a reasoning model launched by DeepSeek. Before producing the final answer, the model first outputs a chain-of-thought section to improve the accuracy of the final response. The API exposes the reasoning_content from deepseek-reasoner to users for viewing, display, and distillation.

💡 Request Example

Basic text chat ✅

curl https://4All API地址/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $4All API_API_KEY" \
  -d '{
    "model": "deepseek-reasoner",
    "messages": [
      {
        "role": "user",
        "content": "9.11 and 9.8, which is greater?"
      }
    ],
    "max_tokens": 4096
  }'

Response example:

{
  "id": "chatcmpl-123",
  "object": "chat.completion",
  "created": 1677652288,
  "model": "deepseek-reasoner",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "reasoning_content": "让我一步步思考:\n1. 我们需要比较9.11和9.8的大小\n2. 两个数都是小数,我们可以直接比较\n3. 9.8 = 9.80\n4. 9.11 < 9.80\n5. 所以9.8更大",
      "content": "9.8 is greater than 9.11."
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 10,
    "completion_tokens": 15,
    "total_tokens": 25
  }
}

Streaming response ✅

curl https://4All API地址/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $4All API_API_KEY" \
  -d '{
    "model": "deepseek-reasoner",
    "messages": [
      {
        "role": "user",
        "content": "9.11 and 9.8, which is greater?"
      }
    ],
    "stream": true
  }'

Streaming response example:

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"让我"},"finish_reason":null}]}

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{"reasoning_content":"一步步"},"finish_reason":null}]}

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{"reasoning_content":"思考:"},"finish_reason":null}]}

// ... more chain-of-thought content ...

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{"content":"9.8"},"finish_reason":null}]}

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{"content":" is greater"},"finish_reason":null}]}

// ... more final answer content ...

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"deepseek-reasoner","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

📮 Request

Endpoint

POST /v1/chat/completions

Authentication method

Include the following in the request header for API key authentication:

Authorization: Bearer $4All API_API_KEY

Where $DEEPSEEK_API_KEY is your API key.

Request body parameters

messages

• Type: array
• Required: Yes

The list of messages containing the conversation so far. Note that if you pass reasoning_content in the input messages sequence, the API will return a 400 error.

model

• Type: string
• Required: Yes
• Value: deepseek-reasoner

The model ID to use. Currently only deepseek-reasoner is supported.

max_tokens

• Type: integer
• Required: No
• Default: 4096
• Maximum: 8192

The maximum length of the final answer, excluding the chain-of-thought output. Note that chain-of-thought output can reach up to 32K tokens.

stream

• Type: boolean
• Required: No
• Default: false

Whether to use streaming responses.

Unsupported parameters

The following parameters are currently not supported:

• temperature
• top_p
• presence_penalty
• frequency_penalty
• logprobs
• top_logprobs

Note: To remain compatible with existing software, setting temperature, top_p, presence_penalty, or frequency_penalty will not produce an error, but they will also not take effect. Setting logprobs or top_logprobs will produce an error.

Supported features

• Chat completion
• Prefix continuation in chat (Beta)

Unsupported features

• Function call
• JSON output
• FIM completion (Beta)

📥 Response

Successful response

Returns a chat completion object, or a streamed sequence of chat completion chunk objects if the request is streamed.

id

• Type: string
• Description: Unique identifier for the response

object

• Type: string
• Description: Object type, value is “chat.completion”

created

• Type: integer
• Description: Response creation timestamp

model

• Type: string
• Description: The model name used, value is “deepseek-reasoner”

choices

• Type: array
• Description: Contains the generated response options
• Properties:
• index : Option index
• message : Message object containing the role, chain-of-thought content, and final answer role : Role, value is “assistant” reasoning_content : Chain-of-thought content content : Final answer content
• finish_reason : Completion reason

usage

• Type: object
• Description: Token usage statistics
• Properties:
• prompt_tokens : Number of tokens used by the prompt
• completion_tokens : Number of tokens used by the completion
• total_tokens : Total number of tokens

📝 Context concatenation notes

During each round of conversation, the model outputs chain-of-thought content (reasoning_content) and the final answer (content). In the next round, the chain-of-thought content from the previous round is not concatenated into the context, as shown below:

Note

If you pass reasoning_content in the input messages sequence, the API will return a 400 error. Therefore, please remove the reasoning_content field from the API response before making the next API request, as shown in the example below.

Example:

from openai import OpenAI
client = OpenAI(api_key="<DeepSeek API Key>", base_url="https://4All API地址")

# Round 1
messages = [{"role": "user", "content": "9.11 and 9.8, which is greater?"}]
response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=messages
)

reasoning_content = response.choices[0].message.reasoning_content
content = response.choices[0].message.content

# Round 2 - only concatenate the final answer content
messages.append({'role': 'assistant', 'content': content})
messages.append({'role': 'user', 'content': "How many Rs are there in the word 'strawberry'?"})
response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=messages
)

Streaming response example:

# Round 1
messages = [{"role": "user", "content": "9.11 and 9.8, which is greater?"}]
response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=messages,
    stream=True
)

reasoning_content = ""
content = ""

for chunk in response:
    if chunk.choices[0].delta.reasoning_content:
        reasoning_content += chunk.choices[0].delta.reasoning_content
    else:
        content += chunk.choices[0].delta.content

# Round 2 - only concatenate the final answer content
messages.append({"role": "assistant", "content": content})
messages.append({'role': 'user', 'content': "How many Rs are there in the word 'strawberry'?"})
response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=messages,
    stream=True
)