Overview
A Codex plan sets your price and your ceiling, but the meter underneath is credits pegged to API token prices, not a flat message count. Because each message burns credits in proportion to the tokens it consumes, the published caps are ranges — a heavy multi-file task lands you near the low end, quick edits near the high end. Two plans carry published ranges on the pricing page; the rest are governed by credit balance alone.
What does each Codex plan cost?
Codex has seven plans: Free at $0, Go at $8 a month, Plus at $20, Pro at $100 (5x) or $200 (20x), Business at $20 per user per month on annual billing — a $25 monthly-billing price is reported but not shown on OpenAI's pricing page — and Enterprise at custom pricing. Every paid tier meters actual work in token-aligned credits rather than a flat message allowance.
How many messages do you get on each plan?
OpenAI publishes message ranges for two plans only: Plus gets 15–90 local messages per 5-hour window on GPT-5.6 Sol, and Pro 20x gets 300–1,800 per window. The other tiers carry no published message range on the pricing page — their effective ceilings come from credit balances, model choice, and how heavy each task runs.
| Plan | Price | Published messages (Sol) | Notes |
|---|---|---|---|
| Free | $0 | Not published | Credit-metered |
| Go | $8/mo | Not published | Credit-metered |
| Plus | $20/mo | 15–90 / 5h | Local messages on GPT-5.6 Sol |
| Pro 5x | from $100/mo | Not published | Credit-metered |
| Pro 20x | $200/mo | 300–1,800 / 5h | Messages on GPT-5.6 Sol |
| Business | $20/user/mo (annual) | Not published | Monthly-billing price reported at $25, not shown on the pricing page |
| Enterprise | Custom | Not published | Existing orgs moved to credits Apr 23, 2026 |
Source: OpenAI Codex pricing page (now served from learn.chatgpt.com) and the Codex rate card · verified 2026-07-09.
How do Codex credits work?
Since April 2, 2026, Codex bills usage in credits aligned to API token prices instead of counting messages. On GPT-5.6 Sol, 1 million input tokens costs 250 credits, cached input 25, and output 1,500. OpenAI's own estimate puts a typical developer at $100–200 a month, varying with model choice, instance count, automations, and fast mode.
| Token type | Credits per 1M tokens (GPT-5.6 Sol) |
|---|---|
| Input | 250 |
| Cached input | 25 |
| Output | 1,500 |
- The April 2, 2026 switch applied to new and existing Plus, Pro, and Business plans plus new Enterprise contracts; existing Enterprise orgs (including Edu, Health, and Gov) followed on April 23, 2026.
- Top-up credits are purchasable when you run out mid-cycle, per OpenAI's rate-card coverage — the help-center article itself blocked our direct verification fetch on 2026-07-09, so we're flagging that one as reported rather than independently confirmed.
- OpenAI's average-spend estimate of ~$100–200 per developer per month comes from rate-card coverage, with the spread driven by model, instances, automations, and fast mode.
How do I check my usage?
Your account's usage dashboard shows remaining quota for the current window, and since CLI v0.144.0 (July 9, 2026) the terminal also surfaces usage-limit reset credit details directly. For exact burn rates per model, keep OpenAI's official rate card open — the dashboard tells you where you stand while the rate card explains why.
The v0.144.0 addition is on the official changelog, and the dashboard is the same surface referenced in the quota-mismatch bug below — worth knowing both, since they can disagree.
Why does Codex say I hit my limit when I haven't?
The best-documented limits bug is a quota mismatch: the account dashboard shows remaining quota while the CLI and app report "usage limit reached." It's tracked as GitHub issue #30041 on openai/codex. If you hit it, confirm your dashboard balance first, then check our errors hub for current workarounds.
When do Codex limits reset?
The published message ranges are metered per 5-hour window. Beyond that, Sam Altman has pledged a usage-limit reset at every additional million weekly users up to 10 million — and with Codex passing 5 million weekly users on June 2, 2026, five of those pledged resets are still on the table.
That pledge makes user milestones a limits story, which is why we track them on the stats page and cover each reset as it lands on the wire.
Changelog for this page
Every edit to this reference, newest first — that's what living document means here.