Best Codex CLI Alternatives (2026)

OpenAI’s Codex CLI is a fast, local terminal agent, and for teams already paying for ChatGPT, it is hard to beat in terms of convenience. The catch is the part nobody mentions on launch day: you are tied to OpenAI’s models. When a different model tops the benchmarks next month, or your finance team questions a per-token bill, that lock-in starts to feel expensive.

This is a roundup for developers actively evaluating an alternative to Codex CLI. We cover what each tool does, which models it supports, whether it is free or open-source, and who it is best for.

Why look for a Codex CLI alternative?

Codex CLI does one thing cleanly: it runs in your terminal, reads your repo, edits files, runs commands in a sandbox, and validates the result. It’s open-source, supports AGENTS.md and MCP servers, and authenticates via a ChatGPT plan or your own API key. Three pressures still push developers to look elsewhere.

Model lock-in. Codex CLI runs OpenAI models, full stop. If you want Claude Opus for a gnarly refactor, Gemini for a huge context window, or a local model for an air-gapped repo, Codex can’t follow, and the “best” coding model has changed hands several times in the past year.

Pricing math. With an OpenAI API key, gpt-5.3-codex runs $1.75 per million input tokens and $14 per million output tokens, and heavy agentic use burns through output tokens fast. The ChatGPT-plan quota route has its own session limits too, so both reward shopping around.

Workflow shape. Codex is terminal-first and local. If your team wants an agent that picks up Linear tickets, answers a Slack mention, or fixes a Sentry error at 3 am without anyone at a keyboard, a local CLI is the wrong layer for that.

Those three pressures sort the alternatives below into clear buckets.

The best Codex CLI alternatives

Each tool here is a real, current option. We list what makes it distinct, which models it supports, how it is priced, and the kind of developer it suits.

Claude Code

Claude Code is Anthropic’s terminal agent and a frequent destination for developers leaving Codex. It reads your full codebase, writes and applies changes, runs shell commands, manages git, and responds to @claude mentions on GitHub pull requests and issues.

• Models: Anthropic’s Claude line (Sonnet and Opus). No third-party models.
• Pricing: Pro is $23/month billed annually (or $28/month billed monthly), the first tier with Claude Code; Max starts at $140/month. Pay-per-token via the API: Sonnet is $2/$10 per million tokens through August 31, 2026 (then $3/$15), Opus is $5/$25.
• Best for: Developers who want a high-autonomy agent and are comfortable in the Claude ecosystem. It is strong on multi-file refactors and git-heavy work.
• Watch out for: Anthropic models only. There’s no way to reach for a different model without leaving Claude Code.

If you are weighing these two head-to-head, we go deep on the tradeoffs in our Codex vs. Claude Code comparison.

Gemini CLI (and Antigravity)

This one needs a current-events warning: Google discontinued free and consumer access to Gemini CLI on June 18, 2026, with no grace period, moving individuals to Antigravity CLI (invoked with agy), a closed-source Go rewrite that didn’t ship with feature parity and cut the free request allowance sharply.

Gemini CLI itself is still Apache-2.0 and still works, but only through a paid Gemini API key or an enterprise Gemini Code Assist license; Standard and Enterprise Code Assist holders keep access. Google’s own migration notice lays out the transition.

• Models: Google’s Gemini models.
• Pricing: No current free consumer tier. Paid Gemini API key or enterprise Code Assist.
• Best for: Teams already on Google Cloud or holding enterprise Code Assist seats.
• Watch out for: The free tier is gone for consumers, and the closed-source Antigravity replacement doesn’t yet have full feature parity.

Cursor

Cursor is an AI code editor first, but it also ships a CLI and background cloud agents that can read files, run tests, and execute terminal commands in their own context.

• Models: Frontier models from multiple providers, selectable per request.
• Pricing: Hobby is free, Pro is $20/month, Pro+ is $60, Ultra is $200, and Teams is $40 per seat, on a credit-based usage model.
• Best for: Developers who want an editor and an agent in one app rather than a pure terminal tool. Its background cloud agents are the closest analog to Codex’s autonomous mode.
• Watch out for: Credit-based pricing can get expensive fast if you lean on background agents heavily.

We cover the editor side in more depth in our Cursor alternatives guide.

Aider

Aider is one of the oldest terminal coding agents, shipping command-line AI pair programming before most of this list existed. Written in Python, it maps your codebase, edits files, commits every change to git automatically with a sensible message, and runs linters and tests across more than 100 languages.

• Models: Nearly any LLM, including Claude, GPT, DeepSeek, and local models through Ollama.
• Pricing: Free and open source under Apache-2.0. You pay your model provider directly.
• Best for: Developers who want maximum model flexibility and a mature, widely deployed tool. The automatic git commit per edit is a genuinely different way of working.
• Watch out for: Terminal-only. No GUI and no cloud/background mode, so it doesn’t cover async or multi-repo workflows.

Aider’s own docs are a good first stop for setup.

OpenCode

OpenCode is an open-source coding agent from the team behind SST and one of the most model-flexible options on this list. It reads, writes, and edits files, runs shell commands, surfaces LSP diagnostics, and uses a Plan-then-Build split so you review changes before it touches anything, running in the terminal, as a desktop app, or in an IDE.

• Models: 75+ providers, including Anthropic, OpenAI, Google, GitHub Copilot, and local models. You can authenticate with an existing Copilot or ChatGPT subscription.
• Pricing: Free and open source under the MIT license. You pay your provider.
• Best for: Developers who want the widest range of providers and a privacy-conscious design. OpenCode stores no code or context data, which matters in sensitive environments.
• Watch out for: A newer project with a shorter track record than Aider or Claude Code. Confirm production readiness for your use case.

See OpenCode’s own docs for setup.

Warp

Warp is a full terminal replacement with built-in agents, written in Rust and GPU-accelerated. It bundles a modern terminal, a file editor, a code review panel, and multi-agent orchestration into a single app and can run multiple agents at once, including Claude Code and Codex.

• Models: Latest models from OpenAI, Anthropic, and Google, with bring-your-own-key support on every plan, including the free tier.
• Pricing: Free tier, Build at $20/month with 1,500 monthly credits, and Business at $50 per user per month.
• Best for: Developers who live in the terminal and want one app that combines shell, editor, and agents. It is also a way to keep using Codex while gaining a second agent alongside it.
• Watch out for: It’s a full terminal replacement, not an add-on, so it’s a bigger switching cost if your team is attached to its current shell setup.

Google Jules

Jules is a different kind of tool: an async, cloud-based coding agent rather than a local terminal CLI. Assign it a task, and it works in the cloud against your connected GitHub repo, opening a PR when done. Paid plans get access to the latest Gemini models, starting with Gemini 3 Pro.

• Models: Google’s Gemini models.
• Pricing: Free tier with 15 tasks per day; higher daily limits come with the Google AI Pro and Ultra plans.
• Best for: Developers who want to fire off background tasks and review PRs later, rather than pair with an agent live in the terminal.
• Watch out for: Async-only, with no live pairing, and the free tier’s daily task cap can bottleneck urgent work.

Free and open-source Codex CLI alternatives

If avoiding a subscription or owning your toolchain is the priority, three tools stand out, all of which are genuinely free to run because you only pay your model provider.

Aider (Apache-2.0) is the most mature, with broad model support and automatic git integration. OpenCode (MIT) is the most provider-flexible, with 75+ backends and a desktop app. Goose, originally built by Block, is now governed by the Agentic AI Foundation at the Linux Foundation and ships as a Rust CLI plus a desktop app. Goose is Apache-2.0, supports any LLM across 15+ providers, including local models via Ollama, and uses native MCP to gain real tools rather than just suggesting code.

One caveat: free and open-source isn’t the same as zero-cost. You still pay for tokens, and for serious agentic work, those bills add up. The savings come from no subscription markup and no provider lock-in, not free inference.

Comparison table

Don’t pick one; orchestrate them: the model-agnostic approach

Every “which CLI should I use” debate quietly assumes you have to commit to one agent. You don’t have to.

Model lock-in stings because you tied your workflow to a single agent, so when the best model changes, your workflow has to follow. We invert that. Tembo is a model-agnostic orchestration layer that runs Codex, Claude Code, Cursor, OpenCode, and other agents across your repos, so “best agent this month” stops being a migration and becomes a dropdown. Under the hood, we let you swap the agent and model for each task across Anthropic, OpenAI, AWS Bedrock, and more, so that one vendor’s outage or pricing change doesn’t stall your team.

To be clear about what we are and aren’t: we’re not a drop-in replacement for the Codex binary in your terminal. The tools above are the CLIs; we’re the layer above them, and we fit when the work outgrows a single developer at a single keyboard. That shows up in a few concrete ways:

• Work spans your team, not just a keyboard. Assign a Linear ticket and get it shipped, tag @tembo in Slack for a PR back, or update an API and its client libraries in one coordinated change across repositories.
• Routine work runs on its own. With automations, you schedule recurring jobs or trigger them from webhook events: generate changelogs from commits, post code reviews, triage Sentry errors, or sync docs when code changes.
• It runs where your code lives. Use our cloud or self-host inside your own VPC on a single VM, which matters for regulated teams.

You stay in control throughout. We propose changes, and you approve, reject, or revise them from Linear, Slack, or GitHub. We don’t replace Codex CLI. We just mean you no longer have to pick one agent forever.

How to choose

The right answer depends on which of the three pressures from the top of this post is hurting the most.

If you want to stay in the terminal and just swap models, go to Aider or OpenCode. Both are open source, both run on almost any provider, and both let you keep your local, file-by-file workflow while escaping OpenAI lock-in.

If you want a more autonomous single agent, Claude Code is a strong default, with Cursor close behind if you also want an editor. Pick the model ecosystem you trust for your hardest tasks.

If the real problem is that work shouldn’t depend on a human in the terminal, that’s an orchestration problem, not a CLI problem. A layer like ours lets you keep Codex, Claude Code, or whatever wins next, while moving the trigger for that work into Linear, Slack, and your CI. Our guide on multi-agent orchestration walks through when it pays off, and our full coding CLI tools comparison profiles 15 agents in detail.

Try the model-agnostic approach

Codex CLI is a solid terminal agent, but the moment your work spans models, repos, or teammates, committing to any single CLI becomes the bottleneck.

If the deeper issue is that you don’t want to bet your workflow on a single agent, start with our free tier, which includes 10 free credits for one repository, and upgrade when you’re ready to run Codex, Claude Code, and the rest from one place. Teams evaluating it across an organization can book a demo to see multi-repo orchestration and self-hosting in action.

Frequently asked questions

What is the best free Codex CLI alternative? Aider, OpenCode, and Goose are all free and open source. You only pay your model provider. Aider is the most mature, OpenCode supports the most providers, and Goose is the most extensible through MCP.

Is there a Codex CLI alternative that isn't locked to a single model? Yes. Aider, OpenCode, and Goose each run nearly any LLM, including local models. If you want to orchestrate several agents across repos at once rather than pick a single CLI, we’re model-agnostic and run Codex, Claude Code, and others side by side.

Is Gemini CLI still a good alternative in 2026? Only with a paid Gemini API key or enterprise Code Assist license. Google ended free consumer access on June 18, 2026, and moved individuals to the closed-source Antigravity CLI.

Do I have to stop using Codex CLI to use these tools? No. Tools like Warp can run Codex alongside other agents, and an orchestration layer like ours runs Codex as one of several agents. Switching is rarely all-or-nothing.