The meter

Plan on the cheap seat. Build on the paid one.

A token is a slice of text the model reads or writes. You do not need the math. You need the habit: thinking is cheap in a browser chat. Agents that edit a whole repo are not.

Rough map

What the paid seats actually buy

Numbers below are from official pages or the Lab dossier, verified 2026-09-16. They will move. The habit will not.

ChatGPT / Codex

Official Codex docs: Free $0, Go $8, Plus $20, Pro from $100. Free chat is real. Codex on Free/Go is limited. Do not plan a product in Codex Pro if a browser window would do.

Grok / Grok Build

SuperGrok $30 includes Build on x.ai/pricing. The agent shares a usage pool with chat, Imagine, and the rest. Long agent sessions and video cost more of that pool than a short question.

Claude / Claude Code

Pro $20, Max from $100. Claude Code is on paid plans and shares the same usage pool as chat. Heavy agent days are why Max exists.

GitHub Copilot

Pro $10. Completions stay unlimited on paid plans. Chat, agent, and CLI burn AI Credits. Cheap if you want autocomplete. Not cheap if you treat it like Grok Build all day on Pro.

Rules that save money

The workflow is the budget

The Lab has dated prices. If a badge says stale, believe it.