● Reach
874k/mo
PyPI downloads · aider-chat · 07.2026
Starting at
from $0
the agent itself is free (Apache-2.0); pay the provider for tokens
Surfaces
3+
cli · uv · installer
Team pace
3/week
Average commits · github, 52 weeks
Quick start · aider-install
$ python -m pip install aider-install
$ aider-install▌benchmarks and competitors - analysis in 3.0 ↓
1.0
Should you get it
Use it if you need
- ✓git-first transparency: every edit as a separate commit, rollback and diff through standard git
- ✓model freedom: your own key, cheap for small tasks and frontier for complex ones
- ✓privacy and offline: local models (Ollama, LM Studio) - code never leaves your machine
- ✓working from Russia: the agent itself is open (Apache-2.0), with no country gate
Choose it for git-first transparency and open source: auto-commit every edit, your own key, any model; skip it if you need an actively developed product and top out-of-the-box quality - development has slowed.
Pass by if needed
- ✗top out-of-the-box quality without dealing with keys - that is Claude Code or Cursor
- ✗active development and fresh frontier: almost no releases for a year, models on the site are outdated
- ✗editor, desktop, or web: aider is pure CLI, one surface
- ✗a cloud fleet of parallel sessions - aider is local; that is Devin's genre
The agent itself is free (Apache-2.0). Pay for tokens: use your own provider key (BYOK) or a local model at no cost. There is no subscription or enterprise plan. It works from Russia: the agent has no country gate, but you need an available model.
2.0
Installation by surface
>_Terminal (CLI)→
Aider's only surface and entire purpose: pair programming with a model in the terminal. Builds a repo map from the codebase, edits files, and auto-commits every change to git with a generated message. Python 3.8-3.13.
$ python -m pip install aider-install$ aider-install→Run aider in a git repository folder and provide your model provider key
>_uv / pip→
The same CLI directly through uv or pip - for those who already have Python tooling. One aider binary, no editor or services.
$ uv tool install --force --python python3.12 --with pip aider-chat@latest$ python -m pip install -U --upgrade-strategy only-if-needed aider-chat>_One-liner (curl / PowerShell)→
One-line installation without manual Python: the script installs an isolated aider. Mac/Linux via curl, Windows via PowerShell.
$ curl -LsSf https://aider.chat/install.sh | sh$ powershell -ExecutionPolicy ByPass -c "irm https://aider.chat/install.ps1 | iex"What you need before starting
You need a git repository (aider works on top of git) and access to a model: your own provider key or a local model (Ollama/LM Studio). No separate editor is required - aider lives in the terminal.
● Availability from Russia
The agent itself is open (Apache-2.0): install it through pip, uv, or a curl script with no country restriction. The caveat - models: Anthropic/OpenAI keys require foreign access and a card; local models (Ollama, LM Studio) and any OpenAI-compatible endpoint work from Russia fully locally.
3.0
Rankings and benchmarks
No Aider+model pair on the leaderboards · card reference points
Aider polyglot (own benchmark)
top gpt-5 (high) 88.0% · 11.2025 · aider's own leaderboard: 225 Exercism tasks in 6 languages, aider+model pairs; snapshot 20.11.2025, 2026 frontier models (Fable 5, GPT-5.5) are not covered
Reach
874k downloads/mo · 07.2026 · PyPI aider-chat; 47.2k GitHub stars, 4.7k forks
vibecoding.tech agent rating
category baseline (beta) · 07.2026 · editorial rating: swe-bench 6/16, terminal-bench 12/16; aider's ceiling = the ceiling of the connected model
Who is measured and how it is calculated - AI agent ranking
● agent ≠ model
A score through aider is always an "aider + model" pair, not a single number: aider's ceiling = the ceiling of the connected model. There is no official aider+model pair in AA Index or Terminal-Bench - the card's references are its own: the aider polyglot leaderboard, coverage, and internal rating.
4.0
What it can do: autonomy, techniques, character
Autonomy axis
Aider's autonomy axis leans toward git control, not the cloud: the agent acts independently within a session, but every step becomes a separate git commit - control is not approvals-per-step, but rollback and diff through standard git. There is no fleet in someone else's cloud.
Aider· levels 01-02
01Edits + git review
aider edits files and auto-commits every edit; post-factum control through git diff and rollback
02Auto lint/test
runs lint and tests itself, reads the output, and fixes issues iteratively
03Background / parallel
several sessions in parallel - aider is one terminal session, it does not do this
04Cloud fleet
launching a batch of sessions in someone else's cloud - aider is local, it does not do this
Asks at every stepWorks while your laptop is closed
Gray levels - Aider cannot do this
Team pace · ● live · api.github.com
In a year 156 commitsPeak 38/wkCurrent week 0
Weekly commits, 52 weeks - is the project alive at a glance
5.0
Pricing, limits, and access from Russia
Plans
One: the agent itself is $0 (Apache-2.0); pay the provider for tokens
Aider has no plans: no subscription, per-seat fee, or enterprise plan. The only cost is inference for the connected model at its provider's rate (BYOK). Local models (Ollama, LM Studio) - no bill. There is no fixed formula of "message = $X": pricing varies by model and volume.
aider.chat
2026-07-08
Free access
Yes: the entire agent is free (Apache-2.0)
It is not a trial that is free, but the entire tool. The only cost is model tokens: your own provider key or a local model (Ollama/LM Studio) - then there is no bill at all. Aider works out of the box with a local model at no cost.
aider.chat
2026-07-08
Limits
No hard limits; cost = model tokens
The agent has no quotas - usage is determined by the connected model. BYOK: pay the provider at its rate. Local model: the limit is your hardware, with no bill. There is no official formula of "message = $X": pricing varies by model and volume.
aider.chat
2026-07-08
● limits expire
All vendors in 2026 rebuilt billing around token credits; limit figures change every week. Each line above has a verification date - do not trust figures without a date, including ours.
6.0
Risks: privacy and control
Code privacy
Local tool; analytics are opt-in, code is not collected
Code and files stay on your machine. Analytics are off by default (opt-in), collecting only anonymous statistics (which models, edit formats, errors) and, according to the vendor, "never collects code, chat messages, keys, or personal data." Disable permanently: aider --analytics-disable. BYOK caveat: while working, prompts and code go to the connected model provider (under your contract); with a local model, they go nowhere.
Aider docs: analytics
2026-07-08
Availability from Russia
Works: no country gate for the agent
There is no direct country-based ToS ban (Apache-2.0): pip, uv, and the curl script install aider without country restrictions. The caveat - models: Anthropic/OpenAI keys require foreign access and a card; local models (Ollama, LM Studio) and any OpenAI-compatible endpoint work from Russia entirely locally. Less restrictive than Devin's direct Russia ban.
GitHub Aider-AI/aider (Apache-2.0)
2026-07-08
Open source
Yes: Apache-2.0, 47.2k stars
The github.com/Aider-AI/aider repository is open under Apache-2.0; 47 151 stars, 4 709 forks (08.07.2026). The agent core, repo map, and edit formats are readable and forkable - unlike proprietary Claude Code, Cursor, and Devin. A notable milestone: aider largely writes itself - the latest release claims 88% of new code (the first such release - v0.47.0, 58% vs 42%).
GitHub Aider-AI/aider
2026-07-08
Maintainer
Paul Gauthier; open project under Apache-2.0
Creator and main maintainer - Paul Gauthier (project since May 2023, repo moved from paul-gauthier/aider to Aider-AI/aider). Developed as an open project under Apache-2.0. Company, funding round, and revenue data are not confirmed by primary sources - not included in the card.
GitHub Aider-AI/aider
2026-07-08
7.0
Aider or Claude Code - git-first open source versus out-of-the-box quality
Aider - an open git-first agent (Apache-2.0) with BYOK: your own key, any model, auto-commit every edit. Claude Code - a closed terminal-first agent on Claude models with maximum quality and active development. The split - "transparency through git and open source" versus "quality ceiling and fresh frontier".
Aider
Claude CodeCardgit-first transparency and open source: auto-commit every edit, repo map, any provider
Strength
depth and out-of-the-box quality on Claude models, terminal-first
its own aider+model pair has not been measured in AA Index / Terminal-Bench; aider polyglot own benchmark: top gpt-5 88.0% (11.2025)
Benchmarks
Terminal-Bench 78.9% with Opus 4.8 (06.2026)
has slowed: last release 08.2025, commits in 2026 are rare (last 22.05.2026)
Development
active, regular releases
the agent itself is $0 (Apache-2.0); pay for provider tokens, local model at no cost
Economics
from $17/mo; limits in 5-hour windows + weekly
1: terminal (CLI); many installation paths (pip, uv, curl)
Surfaces
14: CLI, IDE, desktop, web, iOS, CI, Slack, SDK
any: Claude, OpenAI, DeepSeek, local Ollama/LM Studio (the list on the site is outdated)
Models
Claude only (API/Bedrock/Vertex/Foundry)
Apache-2.0, 47.2k stars; largely writes itself (88% of new code)
Openness
no, proprietary CLI
- →Choose Aider for git-first transparency and open source (auto-commit, repo map, your own key); Claude Code - for the out-of-the-box quality ceiling and active development on Claude models.
Other paths
Aider or OpenCodea more modern open source agent (MIT), active development and 75+ providers
Aider or Clineopen agent in the editor and terminal (Apache-2.0), if you need an IDE focus
Aider or Codexcloud tasks and terminal already in a ChatGPT subscription, without your own keys
Frequently asked questions
Aider or Claude Code - which should you choose?
A different trade-off. Aider is chosen for transparency through git and model freedom: open source (Apache-2.0), your own key, any model, auto-commits for every change, and repo map - but quality depends on the connected model, and development has slowed down (last release 08.2025). Claude Code delivers maximum out-of-the-box quality on Claude models without configuring providers, but it is a closed tool and limited to Claude models. Need git-first open source and model freedom - Aider; need the highest quality ceiling and active development - Claude Code.
How much does Aider cost and what is its price?
The agent itself is free (Apache-2.0): no subscription, per-seat fee, or enterprise plan. You pay only for model tokens - your own provider key (BYOK). With a local model (Ollama, LM Studio), there is no bill at all. There is no fixed formula of "message = $X": usage varies by model and volume (verified 08.07.2026).
Does Aider work in Russia?
Yes. There is no direct country-based ban for the open agent: pip, uv, and the curl script install aider without country restrictions. The caveat - models: Anthropic/OpenAI keys require foreign access and a card, while local models (Ollama, LM Studio) and any OpenAI-compatible endpoint work from Russia entirely locally. This is less restrictive than Devin's direct Russia ban.
Is Aider alive, is the project being developed?
As a tool - alive: 874k PyPI downloads per month, 47.2k stars (12.07.2026), a standard category baseline. But development pace has dropped: latest release v0.86.0 - 09.08.2025 (~11 months ago), commits in 2026 are infrequent (latest 22.05.2026), its own polyglot leaderboard has not been updated since 20.11.2025, and the model list on the site is outdated (Claude 3.7 as recommended). You can use it, but the 2026 frontier has not been added - for active development, look at Claude Code, Codex, or OpenCode.
Is Aider free?
Yes, the agent itself is free and open (Apache-2.0). The money goes not to aider, but to the provider for inference of the connected model. Want no bill at all - connect a local model (Ollama, LM Studio): then aider is entirely free.
What is the aider polyglot leaderboard?
Aider's own benchmark: 225 Exercism tasks in six languages (C++, Go, Java, JavaScript, Python, Rust), measuring a model's ability to edit code through the aider harness. These are "aider + model" pairs, not a ranking of all agents: top on the 20.11.2025 snapshot - gpt-5 (high) 88.0%. Caveat: the leaderboard has not been updated since 20.11.2025, so the 2026 frontier is not reflected in it.
Which is correct: Aider or aider?
The canonical name - Aider (Paul Gauthier's open source project, aider.chat). In Russian, it is written as "aider" - the card can be found under both variants. The PyPI package is called aider-chat, the repository - github.com/Aider-AI/aider.
Watch and read on the topic
Entity · aider
Facts · 14, each with a source and date
Card editorial review · 2026-07-12
Facts verified · 2026-07-08
Where to go next
Rating updates
New cards, price and limit changes, ranking shifts - by email, no spam.