Aider: Git-Native AI Pair Programmer

Overview

Aider is an open-source AI pair programmer for the terminal. Every edit it makes is a git commit: you always have a clean diff, a reviewable history, and a one-command rollback. There is no subscription and no cloud agent layer -- you bring your own API key (Anthropic API, OpenAI API, or any OpenAI-compatible endpoint) and Aider does the rest locally.

Pricing

Aider itself is free and open source (MIT license). You pay only for the model you connect to. At Sonnet 4.6 rates ($3/$15 per million tokens), a productive coding session typically costs $0.50 to $3.00 depending on context size and edit frequency. There is no usage cap or tier.

The polyglot leaderboard

Aider maintains a public benchmark: the Aider polyglot leaderboard, which measures how many coding problems each model can solve correctly when paired with Aider. As of May 2026, GPT-5 (High reasoning) leads at 88%. This is the most model-agnostic benchmark available for agentic coding tasks and is widely cited when comparing frontier models on code.

Git-first design

Every file edit Aider makes is staged and committed with the prompt as the commit message. This creates a natural undo path (git reset HEAD~1), makes review trivial (git show), and means Aider sessions leave no invisible state. Teams that do code review on every AI change find this workflow cleaner than tools that apply edits directly to the working tree.

Where it fits

Aider is the choice for developers who want full control over model selection, prefer the terminal, care about git hygiene, and don't want a SaaS dependency. The trade-off against Claude Code is that Aider is more configuration-heavy upfront (model flags, API keys, architect mode setup) but has no token budget enforced by a vendor tier. Against Cursor, Aider has no GUI but runs cleanly in SSH environments and headless CI.

OpenCode is an emerging Rust-based alternative targeting the same terminal workflow with a TUI layer -- worth watching for teams that want a richer terminal UI than Aider's plain text output.

Field notes

The Aider polyglot leaderboard as of May 2026 shows GPT-5 High at 88%, Claude Opus 4.7 at 84%, and Gemini 2.5 Pro at 81% on the benchmark suite. This is a useful independent signal when choosing a model for coding tasks, separate from vendor claims. [changelog, 2026-05-09]