Which ChatGPT plans include Codex? All of them — here's the breakdown
Since the July 2026 merger, every ChatGPT plan includes Codex usage, Free included. Plan-by-plan limits, the GPT-5.6 tiers, and what it means for agents.
All of them. Since Codex was folded into ChatGPT in July 2026, every plan includes Codex usage — Free ($0) for quick tasks, Go ($8) for lightweight work, Plus ($20) with the full GPT-5.6 family, and Pro (from $100) with 5×–20× higher limits. The plans differ in how much you can use per window and which model tiers you get, not in whether Codex is included.
Why this is suddenly confusing
Until mid-2026, Codex lived partly outside ChatGPT and the answer genuinely depended on your plan. Around July 10, 2026, OpenAI merged Codex fully into ChatGPT, and the inclusion question inverted: Codex is now a capability of every ChatGPT account, metered by your plan’s usage windows. Most outdated answers you’ll find predate the merger.
The plans, as of late July 2026
| Plan | Price | What you get for Codex |
|---|---|---|
| Free | $0 | Quick coding tasks, tight limits |
| Go | $8/mo | Lightweight coding tasks |
| Plus | $20/mo | GPT-5.6 family (Sol, Terra, Luna); roughly 10–100 Sol, 25–200 Terra, or 250–2,000 Luna messages per 5-hour window; credits can extend it |
| Pro 5× | $100/mo | ~5× the Plus windows |
| Pro 20× | $200/mo | ~20× the Plus windows, unlimited Voice |
| Business | $20/user/mo | Plus-level limits, extendable with credits |
The ranges are real, not vagueness on our part: consumption per message varies with how hard the model has to think, so OpenAI publishes windows rather than fixed counts. Prices and limits drift — check OpenAI’s pricing page before deciding, and treat this table as the late-July 2026 snapshot it is.
What this means if you run an agent
The merger quietly changed the economics of running a personal agent. OpenClaw can sign in with a ChatGPT account and draw on these same allowances — which means the addressable answer to “what does the model cost?” is now “possibly nothing beyond the subscription you already pay.” A few practical notes from running agents on these plans: an always-on agent is a heavier consumer than interactive coding, so Plus is the realistic floor for daily agent use (Luna buys the most headroom per dollar); the windows are rolling, so a busy afternoon can pause the agent until the window refreshes — configure a fallback model if silence isn’t acceptable; and connecting from a server needs one security setting enabled first. For the full cost picture including the machine the agent lives on, our OpenClaw cost breakdown is here.