[ENHANCEMENT] Request to support DeepSeek V4 flash and V4 pro models #12173

Activity

1. github-project-automation

   added this to Roo Code Roadmapon Apr 24, 2026
2. github-project-automation

   moved this to New in Roo Code Roadmapon Apr 24, 2026
3. dosubot

   added

   EnhancementNew feature or requestNew feature or request

   on Apr 24, 2026
5. binggee9526 commented on Apr 24, 2026

   

   binggee9526

   on Apr 24, 2026

   More actions

   There are informations of DeepSeek V4 model. @roomote

   1. New model IDs are deepseek-v4-flash and deepseek-v4-pro.
   2. Yes.It's same thinking: { type: "enabled" } parameter as the current reasoner of V4 mode.The parameter reasoning can select high or max.It's deault as high. ref https://api-docs.deepseek.com/

   {
           "model": "deepseek-v4-pro",
           "messages": [
             {"role": "system", "content": "You are a helpful assistant."},
             {"role": "user", "content": "Hello!"}
           ],
           "thinking": {"type": "enabled"},
           "reasoning_effort": "high",
           "stream": false
         }

   3.about price and contex
   for deepseek-v4-flash :

   • pricing:1M INPUT TOKENS (CACHE HIT) | $0.028
     1M INPUT TOKENS (CACHE MISS) | $0.14
     1M OUTPUT TOKENS | $0.28
   • context window size: 1M
   • max output:384K

   for deepseek-v4-pro :

   • pricing:1M INPUT TOKENS (CACHE HIT) | $0.145
     1M INPUT TOKENS (CACHE MISS) | $1.74
     1M OUTPUT TOKENS | $3.48
   • context window size: 1M
   • max output:384K

   ref https://api-docs.deepseek.com/quick_start/pricing
8. JiayunLi-3E8 commented on Apr 24, 2026

   

   JiayunLi-3E8

   on Apr 24, 2026

   More actions

   Opened PR #12175 with my attempt at adding deepseek-v4-flash and deepseek-v4-pro support, including optional thinking mode with reasoning_effort parameter. Feedback welcome! #12175

   It's important to note that the Thinking Mode is enabled by default in the Deepseek API. Therefore, you need to explicitly declare the Thinking Mode on/off switch and the thinking intensity in your request ({"thinking": {"type": "enabled/disabled"}}, {"reasoning_effort": "high/max"}).

   Also, please note that unlike rounds in Thinking Mode where no tool calls are made, rounds in which tool calls are made must return the complete reasoning_content to the API in all subsequent requests. If your code does not return reasoning_content correctly, the API will return a 400 error.

   Below is a simple example code for making tool calls in Thinking Mode:

   import os
   import json
   from openai import OpenAI
   from datetime import datetime
   
   # The definition of the tools
   tools = [
       {
           "type": "function",
           "function": {
               "name": "get_date",
               "description": "Get the current date",
               "parameters": { "type": "object", "properties": {} },
           }
       },
       {
           "type": "function",
           "function": {
               "name": "get_weather",
               "description": "Get weather of a location, the user should supply the location and date.",
               "parameters": {
                   "type": "object",
                   "properties": {
                       "location": { "type": "string", "description": "The city name" },
                       "date": { "type": "string", "description": "The date in format YYYY-mm-dd" },
                   },
                   "required": ["location", "date"]
               },
           }
       },
   ]
   
   # The mocked version of the tool calls
   def get_date_mock():
       return datetime.now().strftime("%Y-%m-%d")
   
   def get_weather_mock(location, date):
       return "Cloudy 7~13°C"
   
   TOOL_CALL_MAP = {
       "get_date": get_date_mock,
       "get_weather": get_weather_mock
   }
   
   def run_turn(turn, messages):
       sub_turn = 1
       while True:
           response = client.chat.completions.create(
               model='deepseek-v4-pro',
               messages=messages,
               tools=tools,
               reasoning_effort="high",
               extra_body={ "thinking": { "type": "enabled" } },
           )
           messages.append(response.choices[0].message)
           reasoning_content = response.choices[0].message.reasoning_content
           content = response.choices[0].message.content
           tool_calls = response.choices[0].message.tool_calls
           print(f"Turn {turn}.{sub_turn}\n{reasoning_content=}\n{content=}\n{tool_calls=}")
           # If there is no tool calls, then the model should get a final answer and we need to stop the loop
           if tool_calls is None:
               break
           for tool in tool_calls:
               tool_function = TOOL_CALL_MAP[tool.function.name]
               tool_result = tool_function(**json.loads(tool.function.arguments))
               print(f"tool result for {tool.function.name}: {tool_result}\n")
               messages.append({
                   "role": "tool",
                   "tool_call_id": tool.id,
                   "content": tool_result,
               })
           sub_turn += 1
       print()
   
   client = OpenAI(
       api_key=os.environ.get('DEEPSEEK_API_KEY'),
       base_url=os.environ.get('DEEPSEEK_BASE_URL'),
   )
   
   # The user starts a question
   turn = 1
   messages = [{
       "role": "user",
       "content": "How's the weather in Hangzhou Tomorrow"
   }]
   run_turn(turn, messages)
   
   # The user starts a new question
   turn = 2
   messages.append({
       "role": "user",
       "content": "How's the weather in Guangzhou Tomorrow"
   })
   run_turn(turn, messages)
9. MaxOTS commented on Apr 26, 2026

   

   MaxOTS

   on Apr 26, 2026

   More actions

   It is due to https://api-docs.deepseek.com/guides/thinking_mode

   https://github.com/dirac-run/dirac works well with the new deepseek models