# Config file for llm: https://github.com/simonw/llm 
# Prices vary for respective API providers. Default prices may not reflect current prices.

# [Structure template]
# {PROVIDER}:
#   {MODEL_NAME}:
#     model_family: {MODEL_FAMILY_NAME}
#     model_alias: {MODEL_API_NAME}
#     model_context_length: {MODEL_WINDOW}
#     model_params:
#       {custom parameter arguments}. For example:
#       max_completion_tokens: 
#       think: (OPTIONAL)
#       temperature:
#       top_p:
#       stop:

# MISTRAL
mistral:
  magistral-medium: # magistral-medium-latest
    model_family: mistral
    model_alias: magistral-medium
    model_context_length: 128000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 2.00
      output: 5.00
  mistral-large: # mistral-large-latest
    model_family: mistral
    model_alias: mistral-large
    model_context_length: 256000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.50
      output: 1.50
  mistral-medium: # mistral-medium-latest
    model_family: mistral
    model_alias: mistral-medium
    model_context_length: 128000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.40
      output: 2.00
  mistral-small: # mistral-small-latest
    model_family: mistral
    model_alias: mistral-small
    model_context_length: 128000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.15
      output: 0.60
  ministral-8b: # ministral-8b-latest
    model_family: ministral
    model_alias: ministral-8b
    model_context_length: 256000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.15
      output: 0.15
  ministral-3b: # ministral-3b-latest
    model_family: ministral
    model_alias: ministral-3b
    model_context_length: 256000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.10
      output: 0.10
  codestral: # codestral-latest
    model_family: codestral
    model_alias: codestral
    model_context_length: 128000
    model_params:
      max_tokens: 8192
      temperature: 0.0
      top_p: 1.0
    cost_usd_mtok:
      input: 0.30
      output: 0.90