Short answer: OpenCode if you want to change models without changing agents, Codex if you already pay OpenAI, Claude Code if you want Anthropic's own tool
Verified August 7, 2026. TL;DR: OpenCode is MIT-licensed, reaches 75+ providers out of the box, and signs in to a ChatGPT Plus, GitHub Copilot or GitLab Duo seat you already pay for. Codex is bundled into every ChatGPT plan, including the $0 one. Claude Code has no free tier and runs Claude models only. Pick by which subscription you already hold.
| OpenCode | Claude Code | Codex | |
|---|---|---|---|
| Made by | Anomaly | Anthropic | OpenAI |
| Models | 75+ providers plus local, built in | Claude only | OpenAI, plus providers you configure |
| Cheapest start | $0, bring your own API key | Paid Claude plan or Console credits | $0 on ChatGPT Free |
| Subscriptions it accepts | ChatGPT Plus, Copilot, GitLab Duo | Any paid Claude plan | Every ChatGPT plan, Free up |
| First-party billing | Go $10/mo, or Zen per request | Included in your Claude plan | Included in your ChatGPT plan |
| Client licence | MIT | None published | Apache 2.0 |
The subscription you already pay for may not travel
OpenCode's docs list ChatGPT Plus, GitHub Copilot and GitLab Duo as subscriptions that work "with zero setup". Claude is the exception, and OpenCode says so plainly: Anthropic "explicitly prohibits" using Pro or Max models this way, and the plugins that once did it were dropped in version 1.3.0. So an OpenAI seat drives two of these three agents. A Claude seat drives one.
Codex is less locked to OpenAI than it looks
Codex's CLI is Apache 2.0 with about 105,000 GitHub stars, and its user-level config accepts a model_providers block, with worked examples for Ollama, Mistral and an LLM proxy. OpenCode is MIT with about 194,000 stars and ships a provider list 75+ long, so the same job needs less config. Anthropic's claude-code repo publishes no open source licence. The gap is convenience, not capability.
The cheapest way in is different for each
Codex is listed at $0 on ChatGPT Free, then $8 on Go and $20 on Plus. OpenCode costs nothing to install and nothing extra to run against a provider key you already have, or $10 a month (first month $5) on OpenCode Go for open models. Claude Code is the outlier: it is included in all paid Claude plans, and there is no free tier.
Three different meters
Claude Code shares one pool with Claude on web, desktop and mobile, resetting on a rolling five-hour window with weekly limits on top. Codex meters five-hour windows too, with weekly limits OpenAI says may apply. OpenCode Go meters dollars: $12 per five hours, $30 a week, $60 a month. Point OpenCode at your own key instead and nobody meters you but your provider.
How to choose, in one line each
Already paying for ChatGPT, start with Codex and pay nothing extra. Want an agent that outlives a vendor switch, or local models behind your own gateway, install OpenCode. Want Claude models, buy a Claude plan and run Claude Code: OpenCode's docs say Anthropic prohibits driving a Pro or Max seat from outside. Wider field: Codex vs Cursor vs Claude Code, Gemini CLI vs Claude Code vs Codex, and the Codex, Claude Code and Copilot breakdown.
Top comments (0)