DeepSeek¶

DeepSeek offers fast, cost-effective reasoning models. Both V4-Flash and V4-Pro support extended context (1M tokens) and structured reasoning.

Provider file: ~/.codefreedom/proxy/config/providers/deepseek.yaml

Environment Variables¶

Variable	Description	Required
`DEEPSEEK_API_KEY`	API key from DeepSeek	Yes
`DEEPSEEK_BASE_URL`	API base URL	No (default: `https://api.deepseek.com`)

Models¶

Model	Context	Max Output	Reasoning	Input $/1M	Output $/1M
DeepSeek-V4-Flash	1,000,000	384,000	Yes	$0.14	$0.28
DeepSeek-V4-Pro	1,000,000	384,000	Yes	$1.74	$3.48

Configuration¶

model_list:
  - model_name: DeepSeek/DeepSeek-V4-Flash
    litellm_params:
      model: deepseek/deepseek-v4-flash
      api_base: os.environ/DEEPSEEK_BASE_URL
      api_key: os.environ/DEEPSEEK_API_KEY
      timeout: 300
      drop_params: true
    model_info:
      id: "deepseek-openai-deepseek-v4-flash"
      db_model: false
      supports_reasoning: true
      mode: chat
      context_window: 1000000
      max_tokens: 1000000
      max_input_tokens: 616000
      max_output_tokens: 384000
      supports_system_messages: true
      supports_native_streaming: true
      supports_vision: false
      input_cost_per_token: 0.00000014
      output_cost_per_token: 0.00000028
      supported_openai_params:
        - tools
        - tool_choice
        - parallel_tool_calls
        - max_tokens
        - max_completion_tokens
        - stream
        - stream_options
        - temperature
        - top_p
        - stop
        - thinking
        - reasoning_effort
        - response_format

See the recipe YAML for the full file with both models.

Enabling¶

Uncomment model entries in deepseek.yaml.
Ensure providers/deepseek.yaml is in the include list in config.yaml.
Set DEEPSEEK_API_KEY in ~/.codefreedom/.env.proxy.secrets.
Restart the proxy.