Aider: Model foobar: Unknown context window size and costs, using sane...

What it means

Aider does not recognize the model name, so it does not know the context window size or token costs. It falls back to an unlimited context window and assumes the model is free. The docs say this is usually not a significant problem and you can probably ignore it. When the name looks like a typo, aider adds "Did you mean one of these?" with similar model names.

Causes

• Specifying a model aider has never heard of (official docs).
• A typo in the model name (official docs: "Did you mean?").
• For Ollama models in Nov 2024, a litellm bug caused spurious warnings (issue #2318).

How to fix it

1. 1
   Fix 1.

   Register the context window and costs in a .aider.model.metadata.json file in your home directory, the git repo root, the launch directory, or pass it with --model-metadata-file.

   JSON

   Copy

   {
       "deepseek/deepseek-chat": {
           "max_tokens": 4096,
           "max_input_tokens": 32000,
           "max_output_tokens": 4096,
           "input_cost_per_token": 0.00000014,
           "output_cost_per_token": 0.00000028,
           "litellm_provider": "deepseek",
           "mode": "chat"
       }
   }

2. 2
   Fix 2.

   Silence model warnings with --no-show-model-warnings (added in v0.30.0).

   Terminal

   Copy

   aider --no-show-model-warnings

3. 3
   Fix 3.

   For the Ollama-specific litellm bug in issue #2318, the maintainer said litellm released a fixed version and suggested upgrading litellm.

   Terminal

   Copy

   pip install -U litellm