config.yaml

model_list:
  - model_name: azure_gpt-5.5
    litellm_params:
      model: azure/azure_gpt-5.5
      api_base: os.environ/AZURE_RESPONSES_OPENAI_BASE_URL
      api_version: preview
      api_key: os.environ/AZURE_RESPONSES_OPENAI_API_KEY
      stream_timeout: 4000
      merge_reasoning_content_in_choices: true
      drop_params: True
      additional_drop_params: ["stream_options", "temperature", "top_p", "presence_penalty", "frequency_penalty", "logprobs", "top_logprobs", "logit_bias", "max_tokens"]
      reasoning:
        summary: detailed
      # verbosity: low
    model_info:
      mode: responses
      background: True
      supports_reasoning: True
      max_input_tokens: 128000
      max_output_tokens: 64000
      supports_tool_choice: True
      supports_vision: True
      supports_response_schema: True
      supports_prompt_caching: True
      input_cost_per_token: 0.00000125
      output_cost_per_token: 0.000010
  - model_name: azure_gpt-5.3-codex
    litellm_params:
      model: azure/azure_gpt-5.3-codex
      api_base: os.environ/AZURE_RESPONSES_OPENAI_BASE_URL
      api_version: preview
      api_key: os.environ/AZURE_RESPONSES_OPENAI_API_KEY
      merge_reasoning_content_in_choices: true
      drop_params: True
      additional_drop_params: ["stream_options", "temperature", "top_p", "presence_penalty", "frequency_penalty", "logprobs", "top_logprobs", "logit_bias", "max_tokens"]
      reasoning:
        summary: auto
    model_info:
      mode: responses
      background: True
      supports_reasoning: True
  - model_name: azure_gpt-5.4-pro
    litellm_params:
      model: azure/azure_gpt-5.4-pro
      api_base: os.environ/AZURE_RESPONSES_OPENAI_BASE_URL
      api_version: preview
      api_key: os.environ/AZURE_RESPONSES_OPENAI_API_KEY
      merge_reasoning_content_in_choices: true
      drop_params: True
      additional_drop_params: ["stream_options", "temperature", "top_p", "presence_penalty", "frequency_penalty", "logprobs", "top_logprobs", "logit_bias", "max_tokens"]
      reasoning:
        summary: detailed
    model_info:
      mode: responses
      background: True
      supports_reasoning: True
# --------------Other Settings--------------------

litellm_settings:
  # Networking settings
  request_timeout: 4000 # (int) llm request timeout in seconds. Raise Timeout error if call takes longer than 10s. Sets litellm.request_timeout
  num_retries: 0
  # fallbacks: [{ "gemini-1.5-pro": ["gemini-1.5-flash"] }]
  allowed_fails: 1 # cooldown model if it fails > 1 call in a minute.
  cooldown_time: 30 # how long to cooldown model if fails/min > allowed_fails
  drop_params: true

general_settings:
  master_key: os.environ/MASTER_KEY # sk-1234 # [OPTIONAL] Only use this if you require all calls to contain this key (Authorization: Bearer sk-1234)

# router_settings:
#  fallbacks:
#    [
#      { "or/gemini-2.0-flash-exp": ["gg1/gemini-2.0-flash-exp"] },
#      { "gpt-3.5-turbo": ["gemini-1.5-flash"] },
#    ]
#  model_group_alias: { "gpt-4": "gemini-1.5-pro" }
#  routing_strategy: simple-shuffle