Skip to content
Agentic Coding Toolkit

Choosing the Right Model

My practical recommendation is simple: use the most powerful model you can access.

In my experience, the frontier models from OpenAI and Anthropic have both been consistently good. Rather than over-optimizing model choice up front, I get better results by starting with a strong model and only dialing down when the task is clearly simple.

My default recommendation

Section titled “My default recommendation”

If you are unsure what to pick, choose the strongest model available and use the high reasoning setting by default.

That gives a good balance between quality, reliability, and speed, while still leaving room to adjust up or down depending on the task.

Why I prefer stronger models

Section titled “Why I prefer stronger models”

For complex coding tasks, stronger models are usually worth it.

This is especially true when the task requires:

  • understanding a large or unfamiliar codebase
  • connecting information across multiple files or layers
  • following a detailed command or prompt structure
  • making careful architectural or refactoring decisions

I have also found that the stronger variants do a better job following the command and prompt structure encoded in the Agentic Coding Toolkit. That matters a lot, because ACT relies on the model not just to write code, but also to respect the workflow and produce output in the expected format.

Choosing a reasoning level

Section titled “Choosing a reasoning level”

Many modern models offer different reasoning levels such as low, medium, high, or xhigh.

Here is the rule of thumb I use:

  • high: my default for most ACT sessions
  • xhigh: worth using for the hardest tasks, especially when deep codebase understanding is required
  • medium: often enough for straightforward implementation work
  • low: fine for simple, well-scoped coding tasks where speed matters most

In particular, I have found high and xhigh noticeably better for tasks that require deeper analysis of the codebase or more precise adherence to the instructions.

How to change the model variant

Section titled “How to change the model variant”

Press ctrl+t to open the variants menu.

OpenCode model picker showing the ctrl+t variants menu with the high reasoning option selected

So the practical takeaway is: start with the strongest model you can access, use high by default, move to xhigh for the hardest tasks, and drop to medium or low when speed matters more than maximum reasoning.

Section titled “Related”