Skip to main content
A

Aider

3.6(145 reviews)

1 comparison available

About Aider

Aider is an open-source AI pair programming tool for the terminal, created by Paul Gauthier in 2023 and growing to 25,000+ GitHub stars by 2025. Aider's design is git-native and terminal-first — it runs in the command line, reads your codebase, and makes changes that are automatically committed to git with descriptive messages. Unlike IDE-integrated tools (Cursor, Copilot, Continue), Aider works in any development environment without editor plugins, making it the preferred AI coding tool for Vim/Emacs power users, remote development over SSH, and automation pipelines. Aider's benchmark performance is exceptional — it consistently ranks #1 or #2 on SWE-bench (a benchmark measuring ability to solve real GitHub issues), outperforming most commercial AI coding tools including early versions of Cursor. Aider supports 60+ LLMs: Claude 3.7 Sonnet, GPT-4o, DeepSeek, Gemini, and local models via Ollama. Aider's architect mode uses a separate 'reasoning' model to plan changes before a cheaper 'editor' model writes the code — combining Claude's reasoning with DeepSeek for cost efficiency. Aider's /add command adds files to the context window; /run executes shell commands and adds output to context; /test runs tests and feeds failures back to the AI for self-correction. Aider handles large multi-file refactors by applying unified diffs rather than rewriting entire files. Aider's cost efficiency (works with cheap models like DeepSeek V3 at $0.014/1M tokens) makes it the choice for automated issue resolution, nightly coding pipelines, and cost-sensitive teams running AI coding at scale.

#1/#2 on SWE-bench — highest real-world issue-solving accuracyGit-native: auto-commits every AI change with descriptive messagesArchitect mode: reasoning model plans, editor model codes — cost-efficientTerminal-first — works over SSH, in scripts, no IDE required

Frequently Asked Questions

How does Aider compare to Cursor?

Aider for developers who prefer terminal workflows, git-centric development, and automated multi-file changes without switching to a new IDE. Aider's SWE-bench performance is competitive with Cursor for complex multi-file tasks. Cursor for developers wanting a full IDE experience with GUI, real-time completions, and codebase search UI. Many developers use both: Aider for heavy refactoring tasks and automated pipeline work, Cursor for day-to-day development.

What models does Aider work best with?

Aider's benchmark results show Claude 3.7 Sonnet (extended thinking mode) achieves the highest SWE-bench scores. For daily use, Claude 3.5 Sonnet and GPT-4o provide the best quality-to-cost ratio. DeepSeek V3 and Qwen2.5-Coder via Ollama work well for free local usage. Aider's architect mode pairs a strong reasoning model (Claude) for planning with a cheaper model (DeepSeek) for code writing, reducing costs by 60-80% on large tasks.

Does Aider work on large codebases?

Aider uses repository maps — a compressed representation of all files, functions, classes, and their relationships — to provide relevant context without putting the entire codebase in the context window. This allows Aider to work on large repos (100K+ lines) by intelligently selecting which files and symbols to include based on the current task. You can also manually /add specific files to prioritize context.