Aider vs Claude Code: which terminal coding tool should you use in 2026?

Aider is a free, open-source (Apache 2.0) "AI pair programming" tool for the terminal that works with almost any model: you add the files to change, it edits them and commits each change to git. Claude Code is Anthropic's coding agent: it finds the files itself, runs commands and tests, and adds MCP servers, subagents, hooks and permission modes, but it runs only on a Claude plan or API key. As of October 2026 the bigger difference is momentum: Aider's last release was 0.86.2 in February 2026 and its repository has had no commits since May 22, while Claude Code ships updates almost daily. In Stack Overflow's 2026 survey, 65.5% of developers using AI agents used Claude Code and 2.1% used Aider.

Key takeaways

• Aider: free and open source, any model with your own API key, git-first, and you choose the files.
• Claude Code: an autonomous agent with MCP, subagents, hooks and six permission modes, on a Claude plan or API key.
• Cost: Aider costs only the tokens you use with your provider; Claude Code starts with Pro at $20 a month.
• Status: no Aider release since February 2026, no commits since May 2026, and no MCP support.
• Choose Aider for model freedom and tight, file-by-file control; Claude Code for an agent that does more on its own.

What are Aider and Claude Code?

Aider calls itself "AI pair programming in your terminal." Paul Gauthier wrote nearly all of it; it has about 49,400 GitHub stars. You install it with python -m pip install aider-install, add files to the chat with /add, and ask for changes. It builds a "repo map" of your repository's files and key symbols so the model knows what exists outside the files you added.

Claude Code is Anthropic's agent for the terminal, VS Code and JetBrains, a desktop app, the web and your phone. It "reads your codebase, edits files, runs commands," deciding for itself which files to open (Anthropic, how to use Claude Code).

How do they work differently?

• Who picks the files: in Aider, you. In Claude Code, the agent searches and reads what it needs.
• Running commands: Aider offers to run a shell command and asks "Run shell command?" first; it can lint after each edit and run your tests. Claude Code runs commands as part of its loop, governed by permission modes and rules (Claude Code plan mode).
• Modes: Aider has code, ask, architect and help modes; architect mode has one model plan the change and another write it. Claude Code has plan mode, subagents and agent teams (Claude Code subagents).
• Git: "Whenever aider edits a file, it commits those changes," and /undo reverts the last one. Claude Code keeps checkpoints you can rewind, and commits when you ask.
• Editors: Aider's watch mode reacts to comments ending in AI! or AI? that you type in any editor. Claude Code has official VS Code and JetBrains extensions.
• Extensions: Claude Code supports MCP servers, skills, hooks and plugins. Aider has no MCP support: a request has been open since February 2025, and two pull requests adding it were closed without merging.

Which models can each use?

Aider "can connect to most LLMs": Anthropic, OpenAI, Gemini, DeepSeek, OpenRouter, Bedrock, Vertex, and local models through Ollama or LM Studio, each with your own API key (Aider). It can't use a Claude Pro or Max subscription, because Anthropic doesn't let third-party tools offer Claude.ai login; its copy-paste mode works with any web chat by hand.

Claude Code runs Claude models (Opus, Sonnet, Haiku and Fable) through a Claude plan, the API, Amazon Bedrock, Google Cloud or Microsoft Foundry (Claude Code pricing).

How much do Aider and Claude Code cost?

• Aider: free; you pay your model provider per token. Aider's leaderboard shows what a full 225-exercise test run cost: $29.08 with GPT-5 on high reasoning, $1.30 with DeepSeek V3.2.
• Claude Code: Pro at $20 a month ($17 billed annually), Max at $100 or $200, Team seats from $20, or API rates. Across enterprise deployments, Anthropic puts the average at "around $13 per developer per active day" (Anthropic).

With a cheap model, Aider can cost far less than any Claude plan; with a top model on the API, the per-token bill is the same either way.

Is Aider still maintained?

Aider has published no announcement about its future, but the activity has slowed sharply:

• Last release: 0.86.2, uploaded to PyPI on February 12, 2026. The last GitHub release is 0.86.0, from August 9, 2025.
• Last commit: May 22, 2026, adding recent Claude model names; there have been 82 commits in 2026 and none since.
• Its leaderboard: "last updated November 20, 2025," with no 2026 models.
• Its docs: the Anthropic page still names Claude 3.7 Sonnet as the default.

It still works with today's models through your API key, and as Apache 2.0 software it can be forked. But for new projects, an agent with active development is the safer bet.

Which is better? Benchmarks and usage

Aider's own polyglot leaderboard tests models, not tools: its top score is GPT-5 (high) at 88.0% (Aider). On the benchmarks that test whole agents, Aider isn't listed: Terminal-Bench 4.0 has Claude Code with Opus 5.5 first at 64.8%, and Artificial Analysis's coding-agent index has Claude Code with Sonnet 5.5 at 68 (best AI coding agents).

In Stack Overflow's 2026 developer survey, among people using AI agents, "Claude Code (66%) and GitHub Copilot (59%) are the clear leaders"; Aider is at 2.1% (Stack Overflow).

Aider vs Claude Code vs OpenCode, Gemini CLI and Cline

• OpenCode: an open-source agent for any model, MIT licensed with about 212,000 stars and used by 13.6% of the survey's agent users, with a full agent loop and plan and build modes (OpenCode vs Claude Code).
• Gemini CLI: Apache 2.0, but it stopped serving free, Google AI Pro and Ultra accounts on June 18, 2026 (Gemini CLI free tier).
• Cline: an Apache 2.0 agent for VS Code and the terminal that asks before every edit and command (Cline vs Claude Code).

More choices are in Claude Code alternatives.

Which should you choose?

• You want any model, including local ones, and to approve every file: Aider still does that well, and costs nothing itself.
• You want an open-source agent that's actively developed: OpenCode.
• You want the agent that tops the agent benchmarks, with MCP and subagents: Claude Code.

Pairing as a team

Aider's name for what it does, pair programming, is two people at one keyboard. Poly (usepoly.co) brings that to the agent: a team works with one Claude Code or Codex agent in a shared browser room, everyone sees each step live, and any member can approve a change before it lands. Free to start. What is Poly?