Aider has been the gold-standard open-source coding agent since 2023, running its own polyglot benchmark (a 225-problem Exercism-derived suite) and scoring competitively on third-party SWE-bench evaluations. The architecture is elegant: a CLI that sits in your repo, sends repo-aware context to whichever LLM you point it at, applies edits, and auto-commits to git so you have a clean history of every change. Cost is purely your LLM API bill — Aider itself is free. This makes it the best choice for cost-conscious heavy users who want frontier-model output but balk at $200/mo subscriptions: pair Aider with DeepSeek or self-hosted Llama via Ollama for near-frontier performance at near-zero marginal cost. The CLI ergonomics are still a step behind Claude Code's polish (no IDE plug-ins, more flags to remember), but for terminal-native engineers it's the most flexible and economical option in the category.
Aider is an open-source CLI coding agent that lets you pair-program with any frontier LLM (Claude, GPT, DeepSeek, local models via Ollama, etc.) directly in your repository. It edits files, runs tests, makes git commits, and publishes its own influential polyglot benchmark leaderboard and consistently scores well on third-party SWE-bench evaluations.