Claude Code or OpenCode: how the two coding agents differ

OpenCode vs Claude Code: models, pricing, open source licence, terminal and IDE use, MCP, skills, hooks, privacy and team plans, and which agent to pick.

Summarize with
A paper sketch of a screen turning into the finished app on a phone

OpenCode and Claude Code are both AI coding agents that read your repo, edit files and run commands from the terminal. The real difference is who owns the stack. Claude Code is Anthropic's own agent: it runs Claude models only, it's included in paid Claude plans, and its licence is Anthropic's terms, not an open source one. OpenCode is MIT licensed, works with 75+ model providers and local models, and costs nothing itself: you pay whichever model you plug in. Pick Claude Code if you want Claude at its best on a flat monthly plan. Pick OpenCode if you need model choice, local models or a tool you can read and change.

I design and build this site with Claude Code. I haven't used OpenCode on a real project, so everything about OpenCode below comes from its official docs, read on 4 November 2026. Both tools ship new versions every few days.

OpenCode vs Claude Code at a glance

Claude CodeOpenCode
Made byAnthropicThe OpenCode project (anomalyco on GitHub)
Version on npm, 4 Nov 20262.1.317 (published 2 November)1.19.5 (published 28 October); an OpenCode 2 line has its own install
LicenceProprietary, Anthropic's termsMIT
ModelsClaude only, through Anthropic or Amazon Bedrock, Google Cloud and Microsoft Foundry75+ providers, plus local models
Cost of the toolIncluded in Pro, Max, Ultra, Team and Enterprise plans, or pay per token with an API keyFree; you pay the model
Where it runsTerminal, VS Code and Cursor, JetBrains, desktop app, web, SlackTerminal, desktop app, VS Code and forks, Zed and JetBrains through ACP
Project rules fileCLAUDE.md (can read AGENTS.md)AGENTS.md, with CLAUDE.md as a fallback
SkillsSKILL.md folders, Agent Skills standardSKILL.md folders, also reads Claude's skill folders
Hooks and pluginsShell hooks, plugins from marketplacesJavaScript or TypeScript plugins with event hooks
MCPYesYes, local and remote, OAuth handled

What each tool is

Claude Code is Anthropic's agentic coding tool. Its overview describes one engine across the terminal, VS Code (and Cursor), JetBrains, a desktop app, the web, Slack and CI, with your CLAUDE.md, settings and MCP servers shared between them.

OpenCode calls itself "an open source AI coding agent" in its docs. It runs as a terminal interface, a desktop app or an IDE extension, and you connect a model with /connect. One thing to watch: the docs offer both OpenCode 1 and a newer OpenCode 2, which have separate install commands and both use the opencode command, so they can clash on one machine.

Models: one family or any provider

Claude Code runs Claude models, through Anthropic or through Amazon Bedrock, Google Cloud and Microsoft Foundry if your company already buys AI there.

OpenCode's providers page says it supports 75+ LLM providers. It lists sign-in for a GitHub Copilot subscription and for ChatGPT Plus or Pro, API keys for the big model companies, and local models through Ollama, LM Studio and llama.cpp.

The catch for Claude fans: you can use Claude in OpenCode with an Anthropic API key, but not with your Claude Pro or Max plan. OpenCode's providers page says plugins that do this exist and that "Anthropic explicitly prohibits this." Anthropic's legal and compliance page agrees: subscription sign-in is for Claude Code and Anthropic's own apps, and third-party tools may not route requests through Free, Pro, Max or Ultra credentials.

Pricing in November 2026

Claude Code has no price of its own. It comes with a paid Claude plan or an API key. On Claude's pricing page on 4 November 2026: Pro is $20 a month billed monthly ($17 a month billed yearly), Max is $100 a month, Ultra is $200 a month, Team is $25 a standard seat or $125 a premium seat billed monthly, and Enterprise lists $20 a seat a month billed yearly with a 20 seat minimum and usage billed on top. The Free plan doesn't include Claude Code. With an API key you pay per token instead.

OpenCode is free to download. What you pay depends on the model:

  • Your own API keys: the provider's normal per-token prices.
  • A subscription you already have: GitHub Copilot or ChatGPT Plus or Pro, as above.
  • Local models: no token bill, only your hardware.
  • OpenCode Go: $10 a month, Go Plus $40 a month, or Go Teams $20 a member a month, with monthly dollar limits per model.
  • OpenCode's pay-as-you-go models: charged per request from credits, with a 4.6% plus $0.31 fee on each credit purchase, and some models free for a limited time.

So which is cheaper? It depends on the model. For Claude, a Claude plan in Claude Code is the only way to pay a flat price, because the subscription can't move across. For cheaper open models, OpenCode wins, because Claude Code can't run them.

Open source and licence

OpenCode's GitHub repo carries the MIT licence: you can read, fork, change and ship the code, which matters if your security team audits tools.

Claude Code's repo has a public issue tracker and plugins, but its LICENSE.md reads "All rights reserved" and points to Anthropic's Commercial Terms. Anthropic's legal page says consumer plans fall under the Consumer Terms and Team, Enterprise and API use under the Commercial Terms. It's free to install, not free to modify.

Terminal, IDE and desktop

Both are terminal first. Claude Code adds its own desktop app with side by side sessions and visual diffs, cloud sessions for long tasks, and handoff from the terminal to the desktop app or a phone.

OpenCode's IDE page covers VS Code, Cursor, Windsurf and VSCodium through an extension, and its ACP page adds Zed, JetBrains IDEs and two Neovim plugins. A GitHub integration runs it in GitHub Actions when someone writes /opencode or /oc in a comment.

Agents, rules and permissions

Claude Code reads a CLAUDE.md at the start of every session (or AGENTS.md if that's what your repo has), keeps an auto memory, and can split work across subagents.

OpenCode's agents page describes two main agents you switch between with Tab: Build, with all tools on, and Plan, which won't edit files except Markdown plans. Its rules page treats AGENTS.md as the main file and falls back to CLAUDE.md, so one AGENTS.md can serve both tools.

MCP, skills, hooks and plugins

MCP. Both connect to MCP servers. OpenCode's MCP page covers local and remote servers and handles OAuth for remote ones. It also warns that every server adds to the context, and that some, like GitHub's, "can easily exceed the context limit". The same is true in any agent: turn on the servers you need, not all of them.

Skills. Claude Code's skills are folders with a SKILL.md file, in ~/.claude/skills for you or .claude/skills for one repo, and they follow the open Agent Skills standard. OpenCode's skills page uses the same SKILL.md format and searches its own folders, then .claude/skills and ~/.claude/skills, then .agents/skills. A skill needs a name that matches its folder and a description.

That's where my own free Claude skills come in. They're seven design skills for accessibility, layout, type, colour, interface copy and polish, built on Jakub Krehel's MIT-licensed better-* skills, and I made them for Claude Code. Each SKILL.md has a matching name and a description under OpenCode's 1,024 character limit, so by OpenCode's docs it should load them from ~/.claude/skills too. I haven't tested that in OpenCode myself. The install steps for Claude Code and the Claude app are in how to install Claude skills, and how I use each one on interface work is in the best Claude skills for design.

Hooks and plugins. Claude Code plugins bundle skills, subagents, hooks and MCP servers into one install from a marketplace, through the /plugin command. Hooks run shell commands at points such as after each file edit. OpenCode's plugins are JavaScript or TypeScript modules that subscribe to events such as tool.execute.before and file.edited, can add custom tools, and can be installed from npm. Claude Code adds Anthropic's official marketplace for you; OpenCode plugins hook into more of the agent's events, and a plugin tool can replace a built-in tool of the same name.

Privacy and where your code goes

In both, your prompts and the files the agent reads go to whichever model you use. What differs is who that is and what they keep.

Claude Code's data usage page says Free, Pro, Max and Ultra users choose whether their data trains future models. With that setting on, data is kept for 5 years; with it off, 30 days. Team, Enterprise and API use isn't used for training, with 30 day retention, and zero data retention for qualifying Enterprise accounts. Telemetry can be switched off with environment variables.

OpenCode's enterprise page says it "does not store any of your code or context data", and that processing happens locally or through direct calls to your provider. So your privacy is your provider's policy, or fully local with a local model. Its pay-as-you-go page notes some free models may use your data while they're free, and the enterprise page recommends turning off share pages.

If code can't leave your network at all, OpenCode with a local model is the only option of the two that does that.

Team use

Claude Code has Team and Enterprise plans, managed settings that can allow or block plugin marketplaces, and HIPAA and zero retention options. Project skills and settings can be committed to the repo for everyone.

OpenCode's enterprise page describes a central config, single sign-on and per-seat Enterprise pricing, with no token charges if you run your own LLM gateway. Go Teams covers smaller groups.

Benchmarks, read with care

Artificial Analysis compares the two on its coding agent tests. When I read it, Claude Code with Claude Opus 5.5 scored 62 on its Coding Agent Index and OpenCode with GLM-5.3 scored 49, while OpenCode cost $5.23 a task against $11.42. Those runs use different models, so they mostly compare the models, not the agents.

Which one should you pick?

Pick Claude Code if:

  • You want Claude models and a flat monthly price.
  • You want the desktop app, cloud sessions, Slack and a plugin marketplace without setup.
  • Your company already has a Claude Team or Enterprise plan.

Pick OpenCode if:

  • You want to switch models per task, or use GPT, Gemini, open models or Copilot.
  • Code must stay on your machines, with a local model.
  • You need an MIT licensed tool you can audit, fork or extend.

Not sure? Keep your rules in AGENTS.md, your skills in ~/.claude/skills and your integrations in MCP servers. All three work in each tool, so switching later is cheap.

For design engineering work, I'd judge the agent by how well it follows your rules files and skills, not by benchmark scores, and keep a person reviewing every screen against the design whichever you choose. That review is the core of my design engineering work: I design the product and build it in production code, with AI tools where they save time and my own review on every change.

Frequently asked questions

Is OpenCode free?

The software is free and MIT licensed. You pay for the model: your own API keys, a GitHub Copilot or ChatGPT subscription, local models at no token cost, or OpenCode Go from $10 a month, according to its docs on 4 November 2026.

Can I use my Claude Pro or Max subscription in OpenCode?

No. OpenCode's providers page says Anthropic explicitly prohibits it, and Anthropic's legal page says third-party tools may not route requests through Free, Pro, Max or Ultra plan credentials. You can use Claude in OpenCode with an Anthropic API key, billed per token.

Is Claude Code open source?

No. Its GitHub repo is public for issues and plugins, but the LICENSE.md says all rights reserved and use is subject to Anthropic's terms. OpenCode is the open source one of the two.

Do Claude skills work in OpenCode?

OpenCode's docs say it loads SKILL.md skills from .claude/skills and ~/.claude/skills as well as its own folders, as long as each skill has a name matching its folder and a description. I use my skills in Claude Code and haven't tested them in OpenCode, so treat that as the documented behaviour.

Can OpenCode run fully offline?

With a local model through Ollama, LM Studio or llama.cpp, its docs say yes for the model calls. Install, updates and any MCP servers that reach the web still need a connection.

Sources and method

I read every source on 4 November 2026 and checked both tools' npm versions the same day. I haven't used OpenCode on a real project, so OpenCode details come only from its official docs. Both tools change fast, so check prices and features before you decide.

  • OpenCode docs, intro and install, date read 4 November 2026: opencode.ai
  • OpenCode docs, providers, date read 4 November 2026: opencode.ai
  • OpenCode docs, Go plans, date read 4 November 2026: opencode.ai
  • OpenCode docs, pay-as-you-go models, date read 4 November 2026: opencode.ai
  • OpenCode docs, skills, rules, agents, MCP servers and plugins, date read 4 November 2026: opencode.ai
  • OpenCode docs, IDE, ACP, GitHub and enterprise, date read 4 November 2026: opencode.ai
  • OpenCode licence on GitHub, date read 4 November 2026: github.com
  • OpenCode on npm (opencode-ai 1.19.5), date read 4 November 2026: registry.npmjs.org
  • Claude Code docs, overview, date read 4 November 2026: code.claude.com
  • Claude Code docs, legal and compliance, date read 4 November 2026: code.claude.com
  • Claude Code docs, data usage, date read 4 November 2026: code.claude.com
  • Claude Code docs, skills and plugins, date read 4 November 2026: code.claude.com
  • Claude Code licence on GitHub, date read 4 November 2026: github.com
  • Claude Code on npm (@anthropic-ai/claude-code 2.1.317), date read 4 November 2026: registry.npmjs.org
  • Claude pricing, date read 4 November 2026: claude.com
  • Artificial Analysis, Claude Code vs OpenCode, date read 4 November 2026: artificialanalysis.ai

If you're choosing an AI coding agent because you need a product designed and built, not only a tool, book a free 30 minute intro call. After it, you get a written scope and one fixed quote.

Tags: Claude Code, OpenCode, AI coding agents, Claude skills, Design engineering

Keep reading

Let’s look at your product.

A free 30 minute call. You leave with a clear next step, whether we work together or not.

Book a free intro call