What a month of Aider actually costs
Aider has a cost component most coding agents do not: the repository map. Before it answers anything, it builds and sends a structured summary of your codebase (files, classes, functions, their relationships) so the model can find things without reading everything.
That is a good design, and it is also a fixed cost you pay on every session, sized by your repository rather than by your question. A 200-file repo pays a different baseline than a 20-file one, and neither depends on what you asked.
https://aicomp.ai/v1).
Create one free →
The two cost parts, and which one is yours to change
Aider's spend splits cleanly:
- Fixed per session: the repo map, plus system instructions. Scales with codebase size.
- Variable: your messages, the files Aider reads, the edits it writes back, and the accumulated turn history.
On small repos the variable part dominates and Aider behaves like any other agent. On large ones the map can be the largest single line item in a short session, you pay it whether you ask one question or twenty. That is the counter-intuitive part: on a big codebase, ten short sessions cost more than one long one.
A realistic day (a few files touched, several reads, maybe a test run) is A day of work on this page's workload is 180k input and 50k output tokens — 3.6M input and 1M output across a 20-day month. Where your own split lands between fixed and variable depends mostly on repo size.
| Model | Gateway rate in / out per 1M tokens | One day | 20 days |
|---|---|---|---|
| claude-sonnet-5 | $1.00 / $5.00 | $0.43 | $8.60 |
| gpt-5 | $0.625 / $5.00 | $0.36 | $7.25 |
| deepseek-v4-pro | $0.66 / $1.98 | $0.22 | $4.36 |
| glm-5.3 | $0.70 / $2.2 | $0.24 | $4.72 |
| qwen3.8-max | $1.00 / $3.00 | $0.33 | $6.60 |
One day = 180k input + 50k output on this page's workload. 20 days = 3.6M input and 1M output. Rates checked 2026-10-27, gateway rates move with upstream promotions, so verify the current number in your dashboard before committing to a budget.
Why input price matters more here than in other agents
Repo maps and file reads are input. Unlike an agent whose cost concentrates in large file rewrites, a chunk of Aider's spend arrives as context before any output is generated. So the input rate is not irrelevant here the way it is on some other workloads But output still costs more per token on most models, so a session that does substantial rewriting still ends up output-dominated.
The practical rule: if your sessions are mostly questions and small edits, compare input rates. If they are mostly rewrites, compare output rates. Aider does enough of both that the answer depends on your own mix.
The levers, in order
Batch work into fewer sessions on large repos. This is the Aider-specific one, and it is the opposite of the advice for chat. Since the map is paid per session, consolidating ten small questions into one session removes nine map sends.
Scope what goes in the map. Aider lets you control which files are included. Pointing it at the parts of the codebase you are actually working on reduces the fixed cost directly, and it improves the answers, because a tighter map gives the model less to confuse itself with.
Add files deliberately. Every file you add to the chat is content sent on subsequent turns, not once. Adding five files to answer one question means carrying all five through the rest of the session.
The part that is easy to miss
Aider works against git, so it can see your history, and it writes commits. That is convenient and it means a session's real cost includes whatever the model read to understand the change. Long-lived sessions that accumulate many file adds are the expensive shape, and nothing in the interface warns you as the context grows.
FAQ
FAQ
Why does Aider cost more on my big repo even when I ask small questions?
Because the repository map is sent per session and scales with the codebase. On a large repo that fixed send can exceed the cost of your actual question, so short sessions on big repos are the most expensive pattern.
Should I use one long session or several short ones?
On a large repo, one long session The map is paid once instead of per session. On a small repo it matters much less, and splitting has the usual benefit of keeping turn history short. Match the pattern to repo size.
Does the repo map change as I edit?
It reflects the files Aider is tracking, so it updates as the session changes them. That is why a long session on a big repo keeps paying a map cost rather than paying it once and forgetting it.
Is Aider cheaper than Cursor or Cline?
They bill differently rather than one being cheaper. Aider's distinctive cost is the per-session repo map; editors that do not send one avoid that fixed cost but often re-send more conversation context per turn. Compare against your own repo size and session style.
How do I reduce cost without changing models?
Scope the map and consolidate sessions. Both reduce the fixed per-session overhead, which on a large codebase is where the money is. Adding files more deliberately reduces the variable side.
Related
- Cline monthly cost — an agent whose cost is concentrated elsewhere
- Coding agent cost per month — the comparison across styles
- Aider with an OpenAI-compatible endpoint — setup
- How to calculate LLM cost — the method behind these numbers
- Cheapest models by workload — pick by shape
- Model price index — every model with a page