SeaWorkCodex Pricing (2026): ChatGPT Plans From Free to $200 & 5-Hour Windows
Codex pricing in 2026: Codex rides your ChatGPT plan — Free, Go, Plus, or Pro at $100/$200 for 5x/20x limits. Credits, 5-hour windows, weekly quotas, and how to track real cost.
Codex is not sold as a separate SKU. As of July 2026, OpenAI Codex access and usage ride your ChatGPT plan — Free ($0) through Pro ($100 for 5× / $200 for 20× relative limits), with Business/Enterprise on seat or flexible credit pricing. You can also run local tasks on an API key and pay standard API token rates.
Codex pricing at a glance
| Plan / path | Typical monthly price (USD) | Codex access (shape) |
|---|---|---|
| Free | $0 | Limited Codex |
| Go | ~$8 | Limited / lightweight coding tasks |
| Plus | ~$20 | Expanded Codex (CLI, IDE, web, iOS; model family access) |
| Pro 5× | ~$100 | ~5× Plus rate limits; max Codex tasks |
| Pro 20× | ~$200 | ~20× Plus rate limits |
| Business | Per seat / flexible credits | Workspace Codex + admin; usage-based add-ons |
| Enterprise/Edu | Quote / flexible | Same agent surfaces; flexible pricing scales with credits |
| API key | Pay per token | CLI/SDK/IDE local tasks; no ChatGPT cloud extras |
Prices and multipliers follow OpenAI’s consumer ladder (chatgpt.com/pricing, openai.com/chatgpt/pricing). Codex-specific limits and credit rates: developers.openai.com/codex/pricing. Confirm the day you buy — OpenAI adjusts plan names and multipliers.
What Plus actually includes for Codex (official): Codex on web, CLI, IDE extension, and iOS; cloud integrations (e.g. code review, Slack); GPT-5.6 family (Sol / Terra / Luna); optional ChatGPT credits to extend past included limits.
What Pro adds: choose 5× or 20× higher rate limits than Plus, plus maximum Codex task headroom and higher-tier ChatGPT features.
How Codex usage actually gets metered
As of July 2026, metering is a stack of windows and units — not a single “messages left” counter.
1. Five-hour windows
Local messages and cloud tasks share a five-hour usage window on consumer plans. OpenAI publishes ranges (not fixed message counts) because cost depends on model, context, tools, and task size. Example shape from the official Plus table (local messages / 5h):
| Model | Local messages / 5h (range) |
|---|---|
| GPT-5.6 Sol | 15–90 |
| GPT-5.6 Terra | 20–110 |
| GPT-5.6 Luna | 50–280 |
Source: Codex pricing — usage limits. Treat ranges as capacity bands, not guarantees.
2. Weekly (and longer) quotas
On top of the rolling window, additional weekly limits may apply. Community reports in July 2026 obsess over this stack: weekly bars at 100% after one heavy task, confusion about whether the 5-hour limit was removed or just hidden, and resets that restore the short window but not the weekly pool. Product UI and policy can lag each other — always trust the in-product usage page for your account.
3. Credits after included limits
When included limits run out, Plus and Pro can buy additional credits instead of upgrading immediately. Business/Enterprise with flexible pricing buy workspace credits. Credits convert token usage into a simpler unit; rate cards are per million tokens, and they differ by model:
| Credits per 1M tokens | Input | Cached input | Output |
|---|---|---|---|
| GPT-5.6 Sol | 125 | 12.5 | 750 |
| GPT-5.6 Terra | 62.5 | 6.25 | 375 |
| GPT-5.6 Luna | 25 | 2.5 | 150 |
OpenAI notes GPT-5.6 usage often averages ~5–40 credits per message. Fast mode and image generation burn included limits / credits faster (images often ~3–5×). Details: credits overview on Codex pricing.
As of July 2026, credit top-ups are commonly referenced at about $40 / 1,000 credits — confirm in the product purchase UI before you buy. OpenAI is also moving toward metering Codex more like API token usage (same underlying units, clearer alignment with platform rates), without a hard cutover date published for every surface.
4. Shared agentic pool
Codex usage is shared with other agentic ChatGPT features once those features bill against the same pool (e.g. ChatGPT Work / Excel agent features on eligible plans). A long “Work” session can steal headroom you thought was reserved for coding.
5. API key escape hatch
Anyone can run extra local tasks with an API key billed at standard API rates. You lose some ChatGPT-cloud goodies; you gain pay-as-you-go with no subscription window — though API rate limits and your organization’s spend limits still apply, so set a budget cap before pointing CI at it.
Is Codex free?
Partially. Free and Go include limited Codex access — enough to try, not enough for a full-time coding agent habit. Plus is the first tier most developers treat as “real” weekly Codex use. Pro is for people who keep hitting Plus ceilings.
So: is OpenAI Codex free? The product is free to try on Free/Go with tight caps. Serious daily use is a paid ChatGPT plan or API spend. There is no unlimited free Codex.
Codex vs Claude Code pricing
| Codex (ChatGPT path) | Claude Code | |
|---|---|---|
| Packaging | Bundled into ChatGPT tiers | Bundled into Claude Pro/Max (or API) |
| Entry paid | Plus ~$20 | Pro $20 |
| Heavy tier | Pro 5× ~$100 / 20× ~$200 | Max 5× $100 / 20× $200 |
| Overage | Credits + API key | API tokens / higher Max |
| Metering feel | 5h window + weekly + credits | 5h + weekly shared Claude limits |
Neither is universally cheaper. OpenAI’s credit rate card makes model choice (Sol vs Terra vs Luna) an explicit cost dial; Anthropic’s API table does the same with Haiku/Sonnet/Opus/Fable. Full Claude breakdown: Claude Code pricing guide. Product-level agent comparison (including Cursor): Cursor vs Claude Code.
Track what you’re actually using
Official surfaces (honest inventory):
- In-product Codex / ChatGPT usage UI — remaining window, weekly bars, credit balance. This is the source of truth for your account.
/usage-style CLI feedback (when available in your Codex client version) — quick check mid-session; not a full historical dashboard.- platform.openai.com usage — API-key spend, not the same as ChatGPT subscription headroom.
What official tools still don’t do well: a live multi-session agent dashboard for local Codex runs (what’s running, stalled, waiting on approval). Community threads in July 2026 are full of “quota hit 100% after one task,” reset confusion, and requests for clearer recovery — which is exactly the gap a session monitor fills.
SeaWork’s Codex usage view is built for that operational layer: watch local Codex agents, sessions, and activity from any device while the agent stays on your machine. Pair it with OpenAI’s official counters for billing truth, and SeaWork for “is this run still healthy?”
FAQ
Is Codex free? Limited free access exists on ChatGPT Free/Go. Daily serious use needs Plus, Pro, Business/Enterprise, or API billing.
How much is Codex per month? Same as your ChatGPT plan if you use the ChatGPT path: commonly $0 / ~$8 / ~$20 / ~$100 / ~$200 for Free / Go / Plus / Pro 5× / Pro 20×, plus optional credits. API path has no seat fee — only tokens.
What happens when I hit Codex limits? The active turn may finish under fair-use rules; new work waits for the window to reset, or you buy credits (Plus/Pro), upgrade, switch to a smaller model, or continue local tasks on an API key.
Do ChatGPT and Codex share limits? Yes for agentic features on the same account/workspace once shared metering applies — Codex, Work, and related agents can draw from one pool. See OpenAI’s Codex pricing notes.
Sources
As of July 2026: