← The general playbook

Cut Aider's API Costs

Aider is already one of the leaner agents; its remaining costs live in the repo map and model choice. Tune --map-tokens, use /clear between tasks, and set --weak-model so commit messages and summaries stop billing at frontier rates.

Where Aider's money actually goes

·The repo map is resent context: brilliant for orientation, and a per-request token cost that scales with repo size.
·Git-centric flow means many small requests (including commit-message generation) that don't need the main model.
·Long chat history accumulates like any agent session.

The Aider-specific tactics

01

Set --weak-model for the cheap chores

Aider natively supports a second, cheaper model for commit messages and history summarization. It's a one-flag change that removes frontier pricing from work that never needed it.

02

Tune --map-tokens to your repo

The default repo-map budget may be oversized for your project, or wasted on a monorepo where you only touch one package. Right-sizing it cuts every single request.

03

/clear between tasks

Aider keeps chat history in context until you clear it. A /clear when switching tasks is free money.

The provider-agnostic tactics (prompt caching, retry budgets, batch APIs) are in the general playbook.

Aider costs: common questions

What is Aider's --weak-model and how much does it save?

It routes auxiliary calls (commit messages, chat summarization) to a cheaper model while the main model handles code. On busy days those auxiliary calls are a meaningful share of requests, at a tenth or less of the price.

Does the Aider repo map cost tokens on every request?

Yes. It's included in context to orient the model. That's usually worth it, but its size is tunable with --map-tokens, and oversizing it on a big repo is a silent per-request tax.

The structural version of all of this: run Aider on a flat monthly price, and the bill stops being a variable to manage. 2-minute Aider setup → · Best models for Aider →