Claude Code vs OpenCode

There is a broad variety of tools you can use for agentic engineering.

Among the most popular tools are Claude Code, OpenCode, Cursor, Visual Studio Code with GitHub Copilot, and Google Antigravity.

Other tools include Codex and Gemini CLI.

The author has used these tools extensively over the last couple of months to test their capabilities.

The author built real projects like buildmygraphic.com, internal projects, and released courses on Cursor, GitHub Copilot, and Claude Code.

The tools can be divided into CLI/TUI tools and IDE tools.

OpenCode has a web UI mode, but the default and the author’s usage is the TUI mode, like Claude Code’s default.

Both OpenCode and Claude Code are used from the command line by default.

Claude Code and OpenCode have IDE integrations, but are primarily optimized for terminal use.

Terminal-first design allows developers of Claude Code and OpenCode to focus on AI agents rather than IDE features, providing more flexibility.

IDEs like Cursor and GitHub Copilot offer auto-completion, while Claude Code and OpenCode do not provide auto-completion powered by those tools.

Claude Code and OpenCode are agentic tools where interaction happens through chat, not inline auto-completion.

Cursor's auto-completion is amazing, and GitHub Copilot's auto-completion is decent but worse than Cursor's.

Claude Code supports Anthropic's Haiku, Sonnet, and Opus models.

Integration of Anthropic models in Claude Code is very good and they work really well in the author’s experience.

The same Anthropic models are available in other tools but run in different harnesses; the Claude Code harness with those models is especially good.

Claude Code now integrates with Olama, allowing use of open models.

Open models available through Olama are currently less capable than Anthropic models and need powerful hardware like large GPU RAM.

OpenCode, Cursor, and VS Code also support Olama and many other models.

Previously, users could access Anthropic models in OpenCode via a Claude Code Max subscription, but Anthropic stopped this.

Claude Code supports fewer AI models than OpenCode, Cursor, and VS Code.

The GPT-5.2 Codex model (or similar) is not available in Claude Code, but is available in OpenCode, Cursor, and GitHub Copilot, and the author likes to use it for complex tasks by switching to OpenCode.

Using Olama with Claude Code could be free, requiring only electricity cost.

You can use a GitHub Copilot subscription in OpenCode.

OpenCode offers a paid subscription (OpenCode Black) with flat-fee usage despite being open source.

Claude Code requires a paid subscription, while a subscription for OpenCode is optional.

OpenCode must be profitable as it is not backed by a large company, whereas Claude Code, Cursor, and GitHub Copilot are heavily subsidized by their companies, potentially offering more usage per plan.

The author prefers the UI and behavior of OpenCode over Claude Code.

OpenCode is an amazing piece of software.

The author switches to OpenCode or Cursor when needing models not available in Claude Code, and prefers IDEs for tasks where he wants to review changes individually.

OpenCode is open source.

OpenCode's open source nature enables transparent, community-driven development with rapid innovation, though all tools innovate quickly and GitHub Copilot sometimes lags.

Claude Code has been a driver of new features, Cursor also to some degree, while GitHub Copilot lagged but catches up.

All these tools support sub-agents and agent skills.

In OpenCode, 'agents' covers both different modes (plan/edit) and sub-agents; in Claude Code, 'agent' refers to a sub-agent and modes are separate.

Memory and rule files are supported by all these tools.

All tools support MCP servers and web fetch for browser access and web requests.

From a feature perspective, there aren't many differences; all tools quickly adopt new mainstream features if one implements them.

Claude Code uses Claude.md and a rules directory; OpenCode supports both agents.md and claude.md; agents.md is an open standard supported by Cursor and GitHub Copilot, but Claude Code does not support agents.md, causing incompatibility.

Cursor has its own rules, GitHub Copilot has its own rules, adding to configuration fragmentation.

Agent skills are stored in tool-specific folders; only the Claude Skills folder is universally supported, reflecting the current age of fragmentation.

The author expects that these configurations will become more standardized in one to two years.

All these tools are decent; the author has not found any that is horrible.

Claude Code works really well, especially with Opus, because it uses the harness from the same company that built the models.

There was a period when GitHub Copilot gave even better results with Anthropic models, but this is subjective and project-dependent.

Code quality depends more on the prompt, context, and use of features like Agents and Skills rather than the specific tool.

All these tools can do a good job if the developer provides good information and reviews the AI's output.