Skip to main content

Choosing a model

AGRO does not choose a model for you. The coding harness that you install selects the model. Use the settings of that harness to change the model. See Harnesses overview.

A public benchmark, for example DeepSWE, compares coding models. A benchmark rank alone does not identify the best model for your task.

Choose a model for a task​

  1. Write down the scope, the risk, the tools, and the context size of the task.
  2. Read the benchmark results for long tasks.
  3. Compare the cost of each candidate model.
  4. Compare the speed of each candidate model.
  5. Remove each model that your provider account does not offer.
  6. Remove each model that your coding harness does not support.
  7. Run each remaining model on one typical task in your sandbox.
  8. Select the model that gives the best result on that task.

When a provider releases a new model, do this procedure again.