OpenCode alternatives: the coding agents that do the same job
Claude Code, Codex CLI, Pi, Hermes, Crush, Aider, Goose, Gemini CLI and more against OpenCode: licence, models, plan sign-in and where each runs.
The tools that do OpenCode’s job, as of November 2026: the labs’ own agents, Claude Code (closed source, Claude models, a Claude plan from $20 a month) and the Codex CLI (Apache-2.0, OpenAI’s models on a ChatGPT plan, local models with --oss); the model-neutral open-source agents, Pi (MIT, deliberately minimal), Crush (Charm’s terminal agent, the continuation of the first project called OpenCode), Aider, Goose and Hermes Agent; and editor-first agents like Cline. If what you want is OpenCode’s mix of open source, any provider, and a ChatGPT or Copilot sign-in, Pi and Hermes come closest. One change catches people out: Gemini CLI lost its free tier on June 18, 2026, and Google now sends individuals to its Antigravity CLI.
Side by side
From each maker’s own site and repository on November 21, 2026. “Plan sign-in” means using a subscription you already pay for instead of an API key.
| Agent | Source | Models | Plan sign-in | Where you use it |
|---|---|---|---|---|
OpenCode 2.3.2 | MIT | 200+ providers, local models, free models with no account | ChatGPT Plus or Pro, Copilot; Claude through your own Claude Code | Terminal, desktop, web, phone |
Claude Code | Closed: “All rights reserved” | Claude only, from Anthropic or a cloud provider | Claude Pro, Max, Ultra | Terminal, VS Code, JetBrains, desktop, web, phone |
Codex CLI 0.171.0 | Apache-2.0 | OpenAI’s; Ollama or LM Studio with --oss; custom providers on the Responses API | ChatGPT plans | Terminal, IDE extension, desktop app, web |
Pi 1.6.0 | MIT | Many providers, OpenRouter included | Through /login | Terminal; print, JSON and RPC modes; SDK |
Hermes Agent 0.23.3 | MIT | Nous Portal, OpenRouter, OpenAI, any OpenAI-compatible endpoint, local | ChatGPT; Copilot by plugin; Claude Max with extra usage | Terminal, desktop app, chat apps |
Crush 0.102.0 | FSL-1.1-MIT (source available) | Many providers, OpenCode Go and Zen keys, Ollama, LM Studio, llama.cpp | Charm’s own Hyper subscription | Terminal |
Aider 0.86.2 | Apache-2.0 | “Almost any LLM, including local models” | None named on its site | Terminal, beside your editor |
Goose 1.60.2 | Apache-2.0 | Over 85 providers, local models built in | None named in its README | Desktop app and CLI |
Gemini CLI 0.70.0 | Apache 2.0 | Gemini only | Gemini Code Assist licences; no free or Google AI plans | Terminal, VS Code companion |
Copilot CLI | Proprietary | Copilot’s model list | A Copilot subscription is required | Terminal |
Cline 4.1.27 | Apache 2.0 | OpenRouter, Anthropic, Ollama, LM Studio and more | None named in its README | VS Code, JetBrains, CLI |
Every one of the terminal agents runs on a remote Linux server over SSH like any other program, kept alive in tmux. What differs is what each offers beyond that: OpenCode serves its own web interface, Claude Code has Remote Control from the Claude app, and Hermes answers in chat apps.
Claude Code
Suits: you mostly use Claude and already pay for a Claude plan. Anthropic’s overview lists the terminal, VS Code and JetBrains, a desktop app (in beta on Ubuntu and Debian), the web, and the Claude app on iOS and Android, with Remote Control to carry a session from any machine to your phone. Pro is $20 a month billed monthly, Max $100 and Ultra $200, all including Claude Code, on Claude’s pricing page.
What you give up: open source (its licence file reads “All rights reserved”) and model choice. Besides Anthropic it runs on Amazon Bedrock, Google Cloud and Microsoft Foundry, and through an LLM gateway set with ANTHROPIC_BASE_URL, but Anthropic documents all of these for Claude models only. It reads a repository’s AGENTS.md when there is no CLAUDE.md, so a project set up for OpenCode carries over. How the two compare in detail, including the one benchmark that ran both on the same model, is in OpenCode vs Claude Code.
Codex CLI
Suits: you have a ChatGPT plan and want an open-source agent built around it. The Codex repository is Apache-2.0; you run codex and choose Sign in with ChatGPT, or use an API key. OpenAI’s pricing page includes Codex in every ChatGPT plan, and Plus at $20 is the first it lists “in the CLI”.
curl -fsSL https://chatgpt.com/codex/install.sh | shCopiedWhat you give up: the breadth of providers. Codex is tuned for OpenAI’s models, though it is less closed than often assumed: its advanced configuration says it “can run against a local ‘open source’ provider such as Ollama or LM Studio when you pass --oss”, and it takes custom providers in config.toml, which must speak OpenAI’s Responses API. OpenCode runs the same GPT models on the same ChatGPT plan, so that pair is the easiest to compare like for like; Codex against Claude Code on three real bugs shows how Codex works on real fixes.
Pi
Suits: you like OpenCode’s openness and want less of everything else. Pi is MIT-licensed, made by Earendil, and its README says “Pi ships with powerful defaults but skips features like sub-agents and plan mode.” You connect “a subscription or API key” with /login, OpenRouter included, and drive it interactively, in print or JSON mode, over RPC, or from its TypeScript SDK. What it does that OpenCode doesn’t is stay small, leaving the rest for you to add, or to ask Pi to build.
What you give up: OpenCode’s built-in plan agent, sub-agents, web interface and free models. Both are covered head to head, installed on the same server, in Pi vs OpenCode, and Pi on its own in what Pi is.
Hermes Agent
Suits: you want one agent for code and for everything else. Nous Research’s Hermes Agent is MIT-licensed, calls itself “The self-improving AI agent”, and answers in Telegram, Discord, Slack, WhatsApp, Signal and email as well as its terminal interface and desktop app. Its providers page covers a ChatGPT plan, GitHub Copilot through a plugin, any OpenAI-compatible endpoint, and local servers such as Ollama, LM Studio, vLLM and llama.cpp. A Claude plan works only on Max with extra usage bought: “Claude Pro subscribers cannot use this path.”
What you give up: a tool built around a code repository. Hermes is a personal agent that also codes, so it carries far more than a coding session needs. What Hermes Agent is sets out the rest.
Crush
Suits: you want a polished terminal interface and broad provider support without OpenCode’s server. Crush is from Charm, and it is where the first project called OpenCode went: that repository, archived in September 2025, says “The project has continued under the name Crush, developed by the original author and the Charm team.” The OpenCode at opencode.ai is a different codebase. Crush’s README lists most hosted providers, takes an OpenCode Go or Zen key, runs local models through Ollama, LM Studio or llama.cpp, uses language servers for context, and installs on Linux, macOS, Windows, Android and the BSDs.
What you give up: a permissive licence today, and plan sign-ins. Crush is under FSL-1.1-MIT, a Functional Source License that “converts to Apache 2.0 or MIT after two years” for each version, so the current release is source-available rather than open source. Its README names no ChatGPT, Claude or Copilot sign-in; the subscription it offers is Charm’s own model service, Hyper.
Aider
Suits: you work in your own editor and want the agent to commit as it goes. Aider is “AI pair programming in your terminal”, Apache-2.0, connects to “almost any LLM, including local models”, and “automatically commits changes with sensible commit messages”, so every edit is a git commit you can undo.
What you give up: plan sign-ins, which its site does not mention, and pace: its latest release on PyPI, 0.86.2, is from February 12, 2026, where every other open-source agent here had a release in November.
Goose
Suits: you want a desktop app as much as a CLI. Goose, now “part of the Agentic AI Foundation (AAIF) at the Linux Foundation”, is Apache-2.0, has a native desktop app for macOS, Linux and Windows beside its CLI, connects to over 85 providers, and can “load local models directly inside goose, without a separate model server.”
What you give up: OpenCode’s ChatGPT and Copilot sign-ins, which its README does not name, and a narrower focus: Goose presents itself as a general agent that goes beyond code.
Gemini CLI, and Antigravity CLI
Older lists put Gemini CLI high for its free tier. That no longer holds. It is still Apache 2.0, but its README now describes it as “maintained for enterprise Gemini Code Assist workflows”, says “No free tier or Google AI plans”, and adds:
Free individual access and Google AI Pro / Ultra (Google One) moved to Antigravity CLI on June 18, 2026CopiedIt runs Gemini models only, and the README warns that its harness is not tuned for Gemini 4. Google’s Antigravity CLI (agy) is now the route for a personal Google account on the Free, Pro or Ultra plan; over SSH it prints a sign-in link to open elsewhere and takes back the code the browser shows. It is Google’s own program under its own terms; its install page shows no open-source licence.
Copilot CLI and the editor agents
GitHub Copilot CLI puts Copilot’s agent in the terminal and needs “an active Copilot subscription”; its licence lets you run it but not modify it. If you already sign OpenCode in with Copilot, it is the same plan in GitHub’s own client.
Agents that live in the editor are a different shape of tool. Cline, Apache 2.0, runs in VS Code and the JetBrains IDEs and now has a CLI, with OpenRouter, Anthropic and local models through Ollama or LM Studio. They suit someone who wants to watch each diff in the editor; on a headless server a terminal agent is the simpler fit.
Before you switch
- Check which OpenCode you are leaving. Version 2 (
@opencode/cli) and version 1 (opencode-ai, 1.19.6 on November 17) differ in features, and a complaint about one may not hold for the other; installing OpenCode on Linux explains which installer gives which. - Count what the model costs, not the tool. All the open-source agents here are free; the bill is the plan or key behind them. What OpenCode’s own Go plan and catalogue cost, and which outside plans it takes, is in OpenCode pricing.
- Your project instructions travel. OpenCode’s
AGENTS.mdis read by Codex, and by Claude Code when there is noCLAUDE.md; which other agents read it is in what AGENTS.md is. - If the confusion is OpenClaw, it is not an OpenCode alternative but a chat assistant with a different job; OpenCode vs OpenClaw separates the two.