# 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:

# max tokens == num_predict
# reasoning level = medium (~8192 budget)
ollama:
  gpt-oss:120b:
    model_family: gpt-oss
    model_alias: gpt-oss:120b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  gpt-oss:20b:
    model_family: gpt-oss
    model_alias: gpt-oss:20b
    model_context_length: 128000
    model_params:
      num_predict: 8192 
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  deepseek-r1:671b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:671b
    model_context_length: 160000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:70b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:70b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:32b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:32b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:14b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:14b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:8b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:7b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:7b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-r1:1.5b:
    model_family: deepseek-reasoner
    model_alias: deepseek-r1:1.5b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-v3:671b:
    model_family: deepseek-chat
    model_alias: deepseek-v3:671b
    model_context_length: 160000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-coder:33b:
    model_family: deepseek-coder
    model_alias: deepseek-coder:33b
    model_context_length: 16000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-coder:6.7b:
    model_family: deepseek-coder
    model_alias: deepseek-coder:6.7b
    model_context_length: 16000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  deepseek-coder:1.3b:
    model_family: deepseek-coder
    model_alias: deepseek-coder:1.3b
    model_context_length: 16000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  magistral:24b:
    model_family: magistral
    model_alias: magistral:24b
    model_context_length: 39000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mistral-large:123b:
    model_family: mistral
    model_alias: mistral-large:123b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mistral-small:24b:
    model_family: mistral
    model_alias: mistral-small:24b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mistral-small:22b:
    model_family: mistral
    model_alias: mistral-small:22b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mistral:7b:
    model_family: mistral
    model_alias: mistral:7b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mistral-medium-3.5:128b:
    model_family: ministral-medium-3
    model_alias: mistral-medium-3.5:128b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  ministral-3:14b:
    model_family: ministral-3
    model_alias: ministral-3:14b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  ministral-3:8b:
    model_family: ministral-3
    model_alias: ministral-3:8b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  ministral-3:3b:
    model_family: ministral-3
    model_alias: ministral-3:3b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  mixtral:8x22b:
    model_family: mixtral
    model_alias: mixtral:8x22b
    model_context_length: 64000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  mixtral:8x7b:
    model_family: mixtral
    model_alias: mixtral:8x7b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  llama3.2:3b:
    model_family: llama3.2
    model_alias: llama3.2:3b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama3.2:1b:
    model_family: llama3.2
    model_alias: llama3.2:1b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  llama3.1:405b:
    model_family: llama3.1
    model_alias: llama3.1:405b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama3.1:70b:
    model_family: llama3.1
    model_alias: llama3.1:70b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama3.1:8b:
    model_family: llama3.1
    model_alias: llama3.1:8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  llama3:70b:
    model_family: llama3
    model_alias: llama3:70b
    model_context_length: 8000
    model_params:
      num_predict: 4096
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama3:8b:
    model_family: llama3
    model_alias: llama3:8b
    model_context_length: 8000
    model_params:
      num_predict: 4096
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  llama2:70b:
    model_family: llama2
    model_alias: llama2:70b
    model_context_length: 4000
    model_params:
      num_predict: 1024
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama2:13b:
    model_family: llama2
    model_alias: llama2:13b
    model_context_length: 4000
    model_params:
      num_predict: 1024
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  llama2:7b:
    model_family: llama2
    model_alias: llama2:7b
    model_context_length: 4000
    model_params:
      num_predict: 1024
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  qwen3-coder:480b:
    model_family: qwen3-coder
    model_alias: qwen3-coder:480b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3-coder:30b:
    model_family: qwen3-coder
    model_alias: qwen3-coder:30b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:122b:
    model_family: qwen3.5
    model_alias: qwen3.5:122b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:35b:
    model_family: qwen3.5
    model_alias: qwen3.5:35b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:27b:
    model_family: qwen3.5
    model_alias: qwen3.5:27b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:9b:
    model_family: qwen3.5
    model_alias: qwen3.5:9b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:4b:
    model_family: qwen3.5
    model_alias: qwen3.5:4b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:2b:
    model_family: qwen3.5
    model_alias: qwen3.5:2b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3.5:0.8b:
    model_family: qwen3.5
    model_alias: qwen3.5:0.8b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  qwen3:235b:
    model_family: qwen3
    model_alias: qwen3:235b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:32b:
    model_family: qwen3
    model_alias: qwen3:32b
    model_context_length: 40000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:30b:
    model_family: qwen3
    model_alias: qwen3:30b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:14b:
    model_family: qwen3
    model_alias: qwen3:14b
    model_context_length: 40000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:8b:
    model_family: qwen3
    model_alias: qwen3:8b
    model_context_length: 40000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:4b:
    model_family: qwen3
    model_alias: qwen3:4b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:1.7b:
    model_family: qwen3
    model_alias: qwen3:1.7b
    model_context_length: 40000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen3:0.6b:
    model_family: qwen3
    model_alias: qwen3:0.6b
    model_context_length: 40000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  qwen2.5:72b:
    model_family: qwen2.5
    model_alias: qwen2.5:72b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen2.5:32b:
    model_family: qwen2.5
    model_alias: qwen2.5:32b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen2.5:14b:
    model_family: qwen2.5
    model_alias: qwen2.5:14b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  qwen2.5:7b:
    model_family: qwen2.5
    model_alias: qwen2.5:7b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  phi4-reasoning:14b:
    model_family: phi4
    model_alias: phi4-reasoning:14b
    model_context_length: 32000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi4-mini-reasoning:3.8b:
    model_family: phi4
    model_alias: phi4-mini-reasoning:3.8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi4:14b:
    model_family: phi4
    model_alias: phi4:14b
    model_context_length: 16000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi4-mini:3.8b:
    model_family: phi4
    model_alias: phi4-mini:3.8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi3.5:3.8b:
    model_family: phi3.5
    model_alias: phi3.5:3.8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi3:3.14b:
    model_family: phi3
    model_alias: phi3:3.14b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi3:3.8b:
    model_family: phi3
    model_alias: phi3:3.8b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  phi:2.7b:
    model_family: phi
    model_alias: phi:2.7b
    model_context_length: 2000
    model_params:
      num_predict: 800
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  command-a:111b:
    model_family: command-a
    model_alias: command-a:111b
    model_context_length: 16000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  command-r:35b:
    model_family: command-r
    model_alias: command-r:35b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  aya:35b:
    model_family: aya
    model_alias: aya:35b
    model_context_length: 8000
    model_params:
      num_predict: 4096
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  aya:8b:
    model_family: aya
    model_alias: aya:8b
    model_context_length: 8000
    model_params:
      num_predict: 4096
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0

  gemma4:31b:
    model_family: gemma4
    model_alias: gemma4:31b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  gemma4:26b:
    model_family: gemma4
    model_alias: gemma4:26b
    model_context_length: 256000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  gemma4:e4b:
    model_family: gemma4
    model_alias: gemma4:e4b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0
  gemma4:e2b:
    model_family: gemma4
    model_alias: gemma4:e2b
    model_context_length: 128000
    model_params:
      num_predict: 8192
      think: True
      temperature: 0.0
      top_p: 1.0
      top_k: 0.0
      repeat_penalty: 1.0