Claude Code vs OpenCode

Claude Code (Opus 4.5) consumed all of its usage limit in just 30 minutes of coding.

Manual coding feels like an ancient relic from 2024.

After using Claude Code extensively during the holidays, the author felt Codex CLI was missing features such as plan mode, LSPs, Vim mode, and sub-agents.

GPT-4.5 with Codex is an extremely powerful model.

Codex CLI is intentionally very simple and lacks advanced tools like LSPs, Vim mode, and sub-agents.

OpenCode now supports logging in with an OpenAI subscription, responding to Anthropic blocking external use of Claude Code subscriptions.

OpenAI fully allows using the ChatGPT Pro subscription with OpenCode.

Using a subscription instead of an API key avoids unlimited pay-per-use costs, which is smart for AI 'addicts'.

OpenCode is an open-source AI coding agent.

Both Codex and Claude Code are also open source, just like OpenCode.

OpenCode lets you choose your model provider, while Codex CLI is coupled to OpenAI and Claude Code is coupled to Anthropic.

OpenCode supports providers like Amazon Bedrock, Google Vertex AI, Cohere, and local Llama cpp models.

OpenCode offers a 'Sen' provider in beta that gives access to various models without signing up with a third party.

The Sen provider’s token costs might be lower due to bulk, but the author suspects they are normal market rates.

OpenCode allows defining permissions for specific built-in commands, a feature not available in Claude Code or Codex CLI.

Users can create custom tools in OpenCode directly, unlike in Codex or Claude Code which require writing a skill and linking a Python script.

OpenCode uses 'agents.md' for rules, while Claude Code uses 'cloud.md', forcing users to rename the file when switching.

OpenCode supports defining multiple agents with their own context, tools, and model, a feature also present in Claude Code but missing from Codex CLI.

Plan mode in OpenCode is a separate agent that first brainstorms and plans, while Codex CLI lacks agents entirely.

OpenCode supports customizable themes and key bindings, features not offered by other coding agents.

Users can define custom slash commands in OpenCode to reuse complex prompts.

Including formatters inside the AI coding tool is a weird conflict of interest; pre-commit hooks would be better.

OpenCode has LSP servers that give the agent language-aware error detection and code navigation.

OpenCode automatically discovers Claude Code skills (named 'cloud') and can execute them without renaming.

OpenCode is far more advanced and feature-rich than Codex CLI.

The default permission mode of OpenCode is more lenient than Codex CLI, which can be a risk.

A simple Flappy Bird clone was built in about one minute using OpenCode, demonstrating the power of coding agents.

Claude Code’s plan mode is still superior, asking interactive questions and being more thoughtful than OpenCode’s.

For those who can afford the $200 Claude Code tier, it remains the best choice; OpenCode does not offer a compelling reason to switch.

OpenCode serves as a much more feature-rich alternative to the basic Codex CLI, adding LSPs, Vim mode, plan mode, and agent definitions.

OpenCode supports using various open-source models and Chinese provider models, but the author has no personal use for them.

The author plans to keep using OpenCode to evaluate whether the GPT-4.5 model performs as well in OpenCode as it does in Codex CLI.