Aider is an Apache 2.0 terminal tool that works with almost any model, including local ones, and auto-commits every change. Cursor is a VS Code fork at $20/mo whose Tab completion is the best in the category. Aider wins on control, model choice, and cost at light use. Cursor wins on the editing hour and on being actively shipped.
- Aider is free software; you pay only your model provider.
- Cursor Pro is $20/mo, and Tab plus Auto mode do not draw credits.
- Aider works with almost any model, including local ones.
- Aider auto-commits every change, which is the best idea in it.
- Aider's last tagged release was v0.86.0 in August 2025. Check that before you standardise on it.
Side by side
Checked against the vendor pages and the GitHub API, August 2026.
| Aider | Cursor | |
|---|---|---|
| Shape | Terminal tool, editor-agnostic | A VS Code fork |
| Licence | Apache 2.0 | Proprietary |
| Price | $0 plus your token spend | Free tier, Pro $20; Pro+ and Ultra prices not published |
| Models | Almost any, including local via Ollama | Several, chosen by Cursor |
| Predictive editing | None | Tab, the best in the category |
| Git integration | Auto-commits every change | Standard editor git |
| Repo awareness | Repo map built from your tree | Indexed codebase |
| Offline / air-gapped | Yes, with a local model | No |
| Runs in CI | Yes, scriptable | A separate CLI exists |
| Team controls | None | Teams $40/user, SSO, analytics |
| Last tagged release | v0.86.0, August 2025 | Continuous |
| GitHub stars | 48,000 | Closed source |
Who is actually cheaper
Free plus tokens is not the same as free, and the comparison has a clean threshold. Cursor Pro is $20 a month and includes a $20 credit pool, unlimited Tab completion, and Auto mode that does not draw on the pool at all. So the question is exactly this: does your Aider token bill come to more or less than $20 a month?
What decides which side of $20 you land on.
| Usage shape | Aider bill | Verdict |
|---|---|---|
| A few sessions a week, a mid-tier model | Low single digits | Aider |
| Daily use, frontier model, large repo map | Comfortably past $20 | Cursor, on price alone |
| A local model on your own hardware | $0 | Aider, decisively |
| Mostly autocomplete while you type | Aider does not do this | Cursor |
| Long agent runs with a big context | Grows fast, no ceiling | Cursor, for the predictable bill |
| Occasional bursts, quiet months | Near zero in a quiet month | Aider |
# recommended installer
python -m pip install aider-install
aider-install
# one-liners
curl -LsSf https://aider.chat/install.sh | sh # macOS, Linux
powershell -ExecutionPolicy ByPass -c "irm https://aider.chat/install.ps1 | iex" # Windows
# or with uv / pipx
uv tool install --force --python python3.12 --with pip aider-chat@latest
pipx install aider-chat
# first run, pick your provider
cd ~/code/your-project
aider --model sonnet --api-key anthropic=<key>The maintenance question
This is the part most comparisons skip, and it is the one that should change your mind. Aider is not abandoned, but its cadence has changed a great deal, and if you are choosing a tool to standardise a team on, that is a fact you want before the decision rather than after.
Aider repository state, read from the GitHub API on 7 August 2026.
| Signal | Value |
|---|---|
| Latest tagged release | v0.86.0, 9 August 2025 |
Last push to main | 22 May 2026 |
| Archived | No |
| Stars | 48,021 |
| Open issues | 1,779 |
| Licence | Apache 2.0 |
Commits are still landing, so the project is alive. But a year without a tagged release, in a category where a new frontier model ships every few weeks, has a specific practical consequence: Aider maintains its model list by hand. The most recent commits at the time of writing were adding newer Claude model names to that list. If you want a model released last month, you may be installing from main, or waiting.
Auto-commit is the best idea in Aider
Aider commits every change it makes, with a generated message. It sounds like a small convenience and it is not. It turns "what did the agent change" from an archaeology exercise into git log, and it makes any individual edit revertable without unwinding the rest of the session.
- Every edit is a commit, so
git revertworks at the granularity of a decision. - Review becomes a log read. You can walk the session in order rather than diffing a pile.
- A bad turn costs one commit, not a session, which changes how much freedom you are willing to give it.
Watch mode means the answer can be both
Aider is editor-agnostic on purpose. Its watch mode monitors your files and acts on AI comments you type in whatever editor you like, which includes Cursor. So the comparison people frame as either/or has a third answer that is genuinely common: Cursor for the editing hour, Aider for the agent loop on your own key.
| Hour of the day | Tool |
|---|---|
| Typing code you are designing | Cursor, for Tab |
| Reading unfamiliar code | Cursor, for the editor |
| A change you can describe in a sentence | Aider, and it commits it |
| Work on a model Cursor does not offer | Aider |
| Anything on an air-gapped machine | Aider, local model |
| A big refactor you want to review as a diff | Cursor |
| A scripted change across many repos | Aider, it scripts cleanly |
Which to pick
| If you | Pick | Why |
|---|---|---|
| Want the typing to be faster | Cursor | Tab has no equivalent in Aider |
| Have a model requirement | Aider | Any provider, including local |
| Work air-gapped or under data rules | Aider | Nothing has to leave the machine |
| Want a predictable monthly bill | Cursor | A subscription has a ceiling |
| Use it a few times a week | Aider | Quiet months cost nothing |
| Are buying for a team | Cursor | Teams billing, SSO, analytics |
| Want the tool to keep pace with new models | Cursor | Someone else maintains that list |
| Want to read and modify the tool itself | Aider | Apache 2.0, Python |
If you are genuinely undecided, the cheap experiment is to run Aider for a fortnight against a key with a spending limit while keeping Cursor on its free tier. At the end you have two numbers that settle it: your token bill, and how often you missed Tab.
Questions people ask
The software is free and Apache 2.0 licensed. You pay your model provider per token, so a light user might spend a couple of dollars a month and a heavy user on a frontier model can pass $20 without noticing.
It is not archived and commits were still landing on main as of 22 May 2026, but the latest tagged release is v0.86.0 from 9 August 2025. In a category where models ship every few weeks, and where Aider maintains its model list by hand, that gap is worth knowing before you standardise on it.
Not entirely. Aider has no predictive editing, which is the single feature people pay Cursor for. It can replace Cursor's agent mode if you want your own model choice and per-token billing.
Yes. Aider is editor-agnostic and its watch mode responds to AI comments you type in any editor, Cursor included. Running both is a real setup rather than a compromise.
Aider, if your token spend stays under about $20 a month or you run a local model, because Cursor Pro is $20 flat and includes a $20 credit pool. Cursor, once you are using a frontier model daily on a large repository.
Yes, including through Ollama, which is the strongest argument for it. That makes air-gapped and data-restricted work possible in a way no subscription editor can match.
It commits every change it makes with a generated message. That makes any individual edit revertable and turns reviewing a session into reading a git log rather than untangling one large diff.
Yes, Cursor ships a command-line agent alongside the editor, with a non-interactive print mode for scripts and pipelines. It does not change the core comparison, which is about who owns the model choice and the meter.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.