Best Claude Code alternatives in 2026: 9 CLI agents

The best Claude Code alternatives in 2026 are OpenAI’s Codex CLI (the closest match, included in every ChatGPT plan), GitHub Copilot CLI (if you already pay for Copilot) and opencode (if you want to pick any model). Aider, Cursor CLI, Amp, Gemini CLI, Qwen Code and Kiro CLI each fit narrower cases. And you may not need to replace Claude Code at all: many developers keep it and run a second agent next to it.

This post answers one question: what can replace Claude Code, and when does each option make sense? If you want a neutral catalog of every terminal agent instead, see the best AI coding CLI tools. Prices and plans below were checked on each vendor’s site and are current as of October 2026.

Why people look for a Claude Code alternative

The usual reasons are practical, not about quality:

• Limits. Claude plans have a rolling 5-hour window and a weekly cap. Hitting them mid-afternoon is the most common trigger. Claude Code usage limits explains how they work.
• Price. Claude Code isn’t in Claude’s free plan. As of October 2026, Pro is $20/month ($17/month billed yearly) and Max starts at $100/month.
• An existing subscription. If your company already pays for ChatGPT, GitHub Copilot or Google Cloud, an agent that uses that plan costs nothing extra.
• Model choice. Some teams want to try other models, use a local model or route work through their own provider.
• A second opinion. A different model reviewing the first one’s work catches different mistakes.

Claude Code alternatives at a glance

The 9 best Claude Code alternatives

1. OpenAI Codex CLI

Codex CLI is the most direct replacement. You run codex in a project, describe the task and it reads, edits and runs commands, just like Claude Code. Its safety model is different: an OS-level sandbox (Seatbelt on macOS, bubblewrap on Linux, a native sandbox on Windows) combined with an approval policy, instead of Claude Code’s permission modes.

npm install -g @openai/codex
codex

• Strengths: strong models, a real sandbox, codex exec for scripts and CI, project rules in AGENTS.md.
• Limits: no equivalent of Claude Code’s hooks and plugin ecosystem; subscription usage has 5-hour and weekly windows too.
• Pricing: as of October 2026, Codex is included in ChatGPT Free, Go ($8/month), Plus ($20/month), Pro (from $100/month), Business, Edu and Enterprise, with higher limits on bigger plans. An OpenAI API key is billed per token.
• Fits: anyone who wants the same workflow with a different model and plan. Claude Code vs Codex goes deeper, and how to use Codex CLI walks through setup.

2. GitHub Copilot CLI

Copilot CLI is GitHub’s terminal agent. It signs in with your GitHub account and offers models from several vendors through your Copilot plan, so one subscription gives you more than one model family. It has interactive, plan and autopilot modes, and it reads AGENTS.md, .github/copilot-instructions.md and even CLAUDE.md.

npm install -g @github/copilot   # Node.js 22+
copilot

• Strengths: no new vendor to approve if your company already uses Copilot; it reads your existing CLAUDE.md; close to GitHub issues and pull requests.
• Limits: usage draws on your plan’s AI credits, so heavy agent work needs a higher plan.
• Pricing: as of October 2026, all Copilot plans include the CLI: Free ($0), Pro ($10/month), Pro+ ($39/month) and Max ($100/month), plus Business and Enterprise seats.
• Fits: teams already paying for Copilot. See Copilot CLI vs Claude Code.

3. opencode

opencode is a terminal agent that isn’t tied to one provider. You connect providers with /connect, pick a model with /models, and switch between a Build agent (edits and runs commands) and a Plan agent (can’t edit) with Tab.

curl -fsSL https://opencode.ai/install | bash
opencode

• Strengths: any provider, local models, a GitHub Copilot sign-in, project rules in AGENTS.md, configuration in opencode.json.
• Limits: you can’t use a Claude Pro or Max subscription with it; Anthropic models need a Console API key billed per token.
• Pricing: free to install; you pay the provider you connect.
• Fits: people who want to choose a model per task, or keep code on local models. Start with how to use opencode.

4. Aider

Aider is one of the oldest terminal coding tools and works differently from Claude Code: you add the files you want changed to the chat, and every edit becomes a git commit, so /undo takes it back.

python -m pip install aider-install
aider-install
aider --model sonnet --api-key anthropic=<key>

• Strengths: works with almost any model, including local ones; ask, code and architect modes; --lint-cmd and --test-cmd to check its own work.
• Limits: less autonomous than Claude Code (you decide which files are in context), and no MCP support in AgentVera.
• Pricing: free to install; you pay your model provider, or nothing with a local model.
• Fits: developers who want small, reviewable commits. See how to use Aider.

5. Cursor CLI

Cursor CLI brings the Cursor agent to the terminal. It has Agent, Plan and Ask modes (switch with /plan, /ask, Shift+Tab or --mode), and it uses your Cursor account.

curl https://cursor.com/install -fsS | bash
agent

On Windows, the docs give a PowerShell install: irm 'https://cursor.com/install?win32=true' | iex.

• Strengths: the same agent you use in the Cursor editor, now scriptable and usable over SSH.
• Limits: usage counts against your Cursor plan; it makes most sense if you already use Cursor.
• Pricing: as of October 2026, Cursor lists Free (limited agent requests), Plus ($8/month), Premium ($30/month), Super ($100/month), Ultra ($200/month) and Teams (from $30/user/month).
• Fits: Cursor users who want a terminal agent too. If you’re weighing the editor itself, see Claude Code vs Cursor and the best Cursor alternatives.

6. Amp

Amp is a multi-model agent: instead of choosing a model, you choose a mode (low, medium, high or ultra) and Amp picks models and tools for that difficulty. It runs on the web and in the terminal, and reads AGENTS.md (falling back to CLAUDE.md when a folder has none).

curl -fsSL https://ampcode.com/install.sh | bash
amp

• Strengths: good defaults without model shopping; reads your existing CLAUDE.md.
• Limits: on Windows the CLI runs through WSL; usage is pay-as-you-go, so costs follow how much you use it.
• Pricing: as of October 2026, Amp has a free tier, and agent usage is paid with credits. For individuals Amp charges the providers’ API prices without a markup.
• Fits: people who want strong results without tuning models and effort themselves.

7. Gemini CLI

Gemini CLI is Google’s terminal agent with Gemini models, GEMINI.md project rules, MCP, checkpoints and several sandbox options. The big change: since June 18, 2026, Gemini CLI no longer serves unpaid and Google One personal accounts; those users were moved to Antigravity CLI.

npm install -g @google/gemini-cli
gemini

• Strengths: long context, strong sandboxing choices (Seatbelt, Docker or Podman, gVisor), fits Google Cloud setups.
• Limits: no longer a free option for individuals.
• Pricing: paid Gemini API key, Vertex AI or a Gemini Code Assist Standard or Enterprise license.
• Fits: companies already on Google Cloud. Details in Gemini CLI vs Claude Code.

8. Qwen Code

Qwen Code is Alibaba’s Qwen-team terminal agent, Apache-2.0 licensed. It started as a Gemini CLI fork and now works with Qwen models and many other providers.

npm install -g @qwen-code/qwen-code@latest   # Node.js 22+
qwen

• Strengths: many providers (QwenCloud, Alibaba ModelStudio, OpenRouter, DeepSeek, local servers), switchable at runtime with /auth.
• Limits: the free Qwen OAuth tier ended on April 15, 2026, so you now need a plan or an API key.
• Pricing: free CLI; you pay through a QwenCloud Token Plan, an Alibaba ModelStudio Coding Plan or an API key.
• Fits: people who want Qwen models or low-cost providers.

9. Kiro CLI

Kiro CLI is the terminal side of AWS’s Kiro, the successor to Amazon Q Developer CLI. You run kiro-cli, sign in in the browser, and use the same subscription as the Kiro IDE.

curl -fsSL https://cli.kiro.dev/install | bash
kiro-cli

• Strengths: a free plan, predictable credit-based pricing, and a natural fit for teams on AWS.
• Limits: credits are shared with the IDE, so heavy use needs a higher plan.
• Pricing: as of October 2026, Free ($0, 50 credits/month), Pro ($20, 1,000 credits), Pro+ ($40, 2,000), Pro Max ($100, 5,000) and Power ($200, 10,000), per user.
• Fits: AWS-centered teams and anyone who wants a capped monthly bill.

How to choose a Claude Code alternative

Ask these in order:

1. What do you already pay for? ChatGPT → Codex. GitHub Copilot → Copilot CLI. Cursor → Cursor CLI. Google Cloud → Gemini CLI. AWS → Kiro CLI. Using a plan you already have is the cheapest switch.
2. Do you need a specific model, or a local one? opencode and Aider let you pick any provider and run local models; Qwen Code also covers many providers.
3. How much autonomy do you want? Codex, Copilot CLI, Cursor CLI, Amp and Kiro run multi-step tasks on their own. Aider keeps you closer to every edit.
4. What does your team need to approve? A tool from a vendor you already use (GitHub, Google, AWS) is easier to get through security review.
5. Test on a real task. Run the same issue through two agents in your own repo and compare the diffs, not the demos.

Switching without losing your setup

Most of what you built around Claude Code carries over if you move it to shared formats first.

One rules file for every agent. Codex, Copilot CLI, opencode and Amp read AGENTS.md. Move your shared rules there and leave a one-line CLAUDE.md so Claude Code reads the same file:

@AGENTS.md

For Gemini CLI, add AGENTS.md to context.fileName in its settings. CLAUDE.md and AGENTS.md covers the details.

Re-add your MCP servers. Each CLI has its own MCP config. For example, adding a server to Codex:

codex mcp add context7 -- npx -y @upstash/context7-mcp

Expect hooks and skills to differ. Claude Code’s hooks, skills and subagents don’t have one-to-one equivalents everywhere. Write down what each hook does (for example, “run the formatter after edits”) and put the instruction in AGENTS.md or the other tool’s own settings.

Commit before every task. Whatever the agent, a git checkpoint before and after makes a bad run cheap to undo.

You don’t have to choose: run several agents side by side

Often the best answer to “what should replace Claude Code?” is “nothing, add a second agent”. Claude Code for a hard refactor, Codex for tests, a cheap model in opencode for docs, and a different model reviewing the result. Each CLI keeps its own login, so nothing stops you from using them together, except that juggling terminal tabs gets messy and two agents in the same folder overwrite each other.

AgentVera is a desktop app for macOS, Windows and Linux that runs Claude Code, Codex, Gemini CLI, Copilot CLI, opencode, Aider, Cursor CLI, Amp, Qwen Code, Kiro and other CLI agents in real terminals, using the installs and logins already on your machine. It is one way to do this; plain terminals with git worktrees work too.

• Workspace: a grid of agents per project, from 1×1 to 5×4, mixing different CLIs. Pick the model and effort per agent.
• Worktrees: start each agent in its own git worktree so they don’t touch each other’s files.
• Code review: have one agent review what another changed before you merge, for example Codex reviewing Claude Code.
• Account switching: when a Claude or Codex limit runs out, the agent moves to the next account in the same conversation (Pro and Team).
• Flows: pass one agent’s reply to another automatically, such as a planner in one CLI and a builder in another (Pro and Team).
• MCP manager: set up MCP servers once and AgentVera passes them to every supported CLI at launch (Aider is the exception).

The grid, worktrees, code review and MCP manager are on the free plan. Download AgentVera and open Claude Code next to one of the alternatives above on the same repo. If you are comparing desktop apps for running agents rather than the agents themselves, the alternatives pages compare AgentVera with tools like Conductor, Warp and Zed.

Wrapping up

If you want one replacement for Claude Code, start with Codex CLI. If you already pay for Copilot, Cursor, Google Cloud or AWS, use that vendor’s CLI first. If you want model freedom or local models, use opencode or Aider. Before you switch, move your rules into AGENTS.md and test two agents on the same real task. And if both turn out to be good at different things, keep both. For a free-only shortlist, see the best free AI coding tools.

More roundups

• Best Claude Code GUI apps (2026)
• Best Cursor alternatives in 2026: 10 tools compared
• Best free AI coding tools (2026): 10 options compared