TL;DR
Codex is included in ChatGPT Plus at $20/month. If you want 6x more capacity, ChatGPT Pro costs $200/month. Teams pay $25/user/month. Enterprise pricing is custom.
Usage resets in 5-hour rolling windows, not monthly. The published ranges (45-225 messages) span a 5x gap because task complexity drives actual consumption more than raw counts. Complex refactors can drain a Plus window in 2-4 tasks.
If you code fewer than 10-15 sessions per month, the API ($0.50-$2.00 per session) is cheaper than a subscription. The free tier exists but is a temporary promo.
Current status (March 2026)
2x rate limit promo is active for Plus and Pro users. GPT-5.3-Codex launched in February 2026 with Pro-tier preview access. The free tier Codex access is a promotional offer with no confirmed end date.
Pricing Tiers
Codex is not a standalone product with its own subscription. Access comes bundled with ChatGPT plans. Here is every tier as of March 2026:
| Plan | Price | Codex Access | Key Additions |
|---|---|---|---|
| Free / Go | $0 (promo) | Limited (temporary) | Basic ChatGPT features only |
| ChatGPT Plus | $20/month | Full: web, CLI, IDE, iOS | Cloud integrations, latest models, 45-225 msgs/5h |
| ChatGPT Pro | $200/month | Full + priority processing | GPT-5.3-Codex preview, 6x higher limits, 300-1,500 msgs/5h |
| ChatGPT Business | $25/user/mo (annual) or $30/mo | Full + larger VMs | Admin controls, SAML SSO, MFA, no training on data |
| ChatGPT Enterprise | Custom (contact sales) | Priority + credits pool | SCIM, EKM, RBAC, audit logs, no fixed rolling limits |
| API Key (usage-based) | Per-token billing only | All models (delayed access) | No cloud features, no ChatGPT integration, pure API |
The Business plan matches Plus on usage limits. The main Business upgrades are larger VM instances for cloud tasks and the security and compliance stack (SSO, MFA, no training on company data). For small teams that hit limits, Pro may make more sense than Business.
Usage Limits
Codex publishes ranges, not fixed numbers. The actual number you get depends on task complexity. Here are the published ranges by plan:
| Plan | Local Messages / 5h | Cloud Tasks / 5h | Code Reviews / week |
|---|---|---|---|
| Plus | 45-225 | 10-60 | 10-25 |
| Pro | 300-1,500 | 50-400 | 100-250 |
| Business | 45-225 (same as Plus) | 10-60 | 10-25 |
| Enterprise | Credits pool (no fixed limit) | Credits pool | Negotiated |
The 2x promo
Since February 2026, OpenAI has been running a 2x rate limit promotion for Plus and Pro users. The upper bound of the published ranges already reflects this promo. If the promo ends, limits drop roughly to half the current highs.
The 5-Hour Rolling Window Explained
Most AI tools reset usage monthly. Codex uses rolling 5-hour windows, which changes the math significantly.
How it works: your window starts when you send your first message. It resets 5 hours later. If you exhaust your allocation before the reset, you wait, or you buy credits to continue. There is no monthly bank that accumulates unused capacity.
Why this matters: a developer who codes in 3-hour sprints can hit their Plus limit on a single complex task and be blocked mid-session. The window resets independently for each user, not at a fixed clock time, so "reset at midnight" strategies do not apply.
November 2025 bug
On November 2, 2025, a bug caused entire 5-hour quotas to drain in as few as 8 queries. OpenAI acknowledged the issue and compensated affected users with $200 in credits. This incident highlighted how opaque the rolling window system is when it malfunctions.
The rolling window also resets separately for local and cloud tasks. Burning your cloud task allocation does not affect your local message count, and vice versa.
Task Complexity Determines Actual Cost
The published ranges are nearly useless without knowing what counts as one task. Here is how task complexity maps to message consumption:
| Task Type | Files Touched | Messages Used | Examples |
|---|---|---|---|
| Simple | 1-2 files | 1-3 messages | Fix a bug, rename a function, add a comment |
| Medium | 3-5 files | 5-15 messages | Add a feature, refactor a module, write tests for a component |
| Complex | 10+ files | 20-50 messages | Architectural refactor, adding a new service, migration |
A Plus user with a 45-message lower-bound window can complete 45 simple bug fixes or roughly 3-4 complex refactors before hitting the limit. At the upper bound (225 messages), that rises to about 225 simple tasks or 15-20 complex ones. Most real workdays land somewhere in between.
Cloud tasks (running in Codex.com sandboxes) use more quota than equivalent local tasks. A cloud task that deploys and tests a change may consume 3-5x the messages of the same change applied locally.
API Pricing
API key access (via OpenAI's API, not chatgpt.com) is billed per token with no monthly subscription. Access was added September 23, 2025. Model availability via API trails subscription access by several weeks.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Best For |
|---|---|---|---|
| gpt-5.1-codex-mini | $0.25 | $2.00 | High-volume, simple tasks where cost matters most |
| codex-mini-latest | $1.50 | $6.00 | Everyday coding tasks, good balance |
| gpt-5.2-codex | $1.25 | $10.00 | Complex reasoning, multi-step planning |
| gpt-5.3-codex | ~$1.75 | ~$14.00 | Highest capability, Pro-tier preview only |
Token efficiency note: coding sessions tend to be output-heavy. Generating 500 lines of code produces roughly 2,500-4,000 output tokens. At codex-mini-latest rates, that is $0.015-$0.024 per generation. A full session involving planning, multiple iterations, and test writing typically totals $0.50-$2.00.
API access does not include cloud features: no Codex.com sandbox, no GitHub integration, no IDE plugin with subscription features. It is raw model access only.
Credits System
When you exceed your plan's rolling-window allocation, you can purchase credits as a pay-as-you-go add-on. Key details:
- Credits are consumed only after your plan's included usage is exhausted
- Credits expire 12 months from purchase date
- No rollover: unused monthly usage does not carry forward
- Credits are billed at standard API rates for your tier
- Enterprise users get a credits pool instead of rolling-window limits
Credits math
One developer reported spending $48 on credits, which was consumed in 8 queries during the November 2025 bug. Under normal conditions, $48 at API rates covers roughly 24-96 typical coding sessions at $0.50-$2.00 each. If you are burning through credits quickly, audit whether cloud tasks are the culprit.
Competitor Pricing Comparison
How Codex pricing stacks up against the major alternatives as of March 2026:
| Tool | Free Tier | Individual/Pro | Team | Enterprise |
|---|---|---|---|---|
| OpenAI Codex | $0 (promo) | $20/mo (Plus) or $200/mo (Pro) | $25-30/user/mo | Custom |
| GitHub Copilot | $0 (2k completions/mo) | $10/mo or $39/mo (Pro+) | $19/user/mo | $39/user/mo |
| Cursor | $0 (limited) | $20/mo | $40/user/mo | Custom |
| Claude Code | $20/mo (capped usage) | $100-200/mo (Max plan) | $125/seat | Custom |
| Windsurf | $0 (25 credits/mo) | $15/mo | $30/user/mo | $60/user/mo |
At the $20/month individual tier, Codex and Cursor are equal on price, but Codex comes bundled with a full ChatGPT subscription. That makes it a better value if you already pay for ChatGPT. If you only want an IDE coding assistant, Windsurf at $15/month or GitHub Copilot at $10/month undercut it.
At the Pro tier ($200/month), Codex competes with Claude Code Max ($200/month). Codex Pro includes 6x more rolling-window capacity. Claude Code Max includes 20x Pro usage. Token efficiency differs by task type, and this comparison deserves its own analysis (see our Codex vs Claude Code deep dive).
Which Plan Should You Pick
Use this decision guide based on your usage pattern:
API key only (no subscription)
Best if: you code fewer than 10-15 sessions per month, you want programmatic access without a GUI, or you are building a product on top of Codex. At $0.50-$2.00 per session, you break even at 10-40 sessions vs Plus.
ChatGPT Plus ($20/month)
Best if: you already have a ChatGPT Plus subscription, you code regularly but not in long daily sprints, and you want IDE integration plus the web interface. The Plus rolling window (45-225 messages) is enough for most developers who do not code for 8+ hours straight.
ChatGPT Pro ($200/month)
Best if: you hit Plus limits regularly, you run complex multi-file refactors frequently, or your time cost of waiting for a window reset exceeds the $180/month price gap. Pro's 6x limits mean it scales to roughly 10x the practical workload of Plus on complex tasks.
ChatGPT Business ($25-30/user/month)
Best if: your team needs data privacy guarantees (no training on company code), SAML SSO, or admin oversight. Usage limits are identical to Plus, so Business is not a capacity upgrade. It is a compliance and control upgrade.
Enterprise (custom)
Best if: you have more than 150-200 seats, need SCIM/EKM/RBAC, or require SLA-backed uptime. The credits-pool model removes rolling window friction entirely, which alone justifies Enterprise for teams running large autonomous coding agents.
When the API Beats a Subscription
This is the calculation most pricing guides skip. API access at $0.50-$2.00 per session beats Plus ($20/month) when you have fewer than 10-40 sessions per month. It beats Pro ($200/month) when you have fewer than 100-400 sessions per month.
The breakeven shifts based on session length. Short sessions ($0.50 average) put the Plus breakeven at 40 sessions per month. Long sessions ($2.00 average) put it at 10 sessions per month.
Who benefits most from the API route:
- Developers who batch work into occasional deep sessions rather than daily use
- Teams building internal tools on top of Codex that need programmatic access
- Developers who already have IDE tooling and do not need the ChatGPT web interface
- Researchers testing Codex capabilities without committing to a subscription
What you lose with API-only: Codex.com cloud sandboxes, GitHub integration, IDE plugin subscription features, and mobile (iOS) access. If those integrations are core to your workflow, a subscription is worth it.
Community Cost Reports
Published pricing ranges are one thing. Real usage patterns are another. Here is what Codex users have reported publicly:
"8 tasks already count as 60% of my weekly cap, and 3 cloud tasks drained the entire 5-hour window."
"2 tasks and I've hit a fresh 5-hour limit. Both were large refactors across 15+ files."
"Most coding sessions cost me $0.50-$2.00 at API rates. If you are not coding every single day, API is cheaper than Pro."
"Spent $48 on extra credits, burned through in 8 queries. Was partly the November bug, OpenAI gave me $200 in compensation."
The pattern: Plus users who run complex tasks hit limits faster than the published upper bounds suggest. The 5-hour window is measured in messages, not time, so a highly iterative session with back-and-forth debugging burns capacity faster than a clean implementation session of equal clock time.
Pricing History
Codex access and pricing has expanded significantly since mid-2025:
| Date | Event |
|---|---|
| April 2025 | Codex CLI launched as free, open-source tool |
| May 2025 | Codex agent launched, Pro and Enterprise only |
| June 3, 2025 | Expanded to ChatGPT Plus users |
| September 23, 2025 | API key access added for all OpenAI API customers |
| November 2, 2025 | Credits bug: entire quotas drained in 8 queries; OpenAI compensated with $200 credits |
| February 2026 | GPT-5.3-Codex launched with Pro preview access; macOS app released; 2x rate limit promo starts |
| March 2026 | 2x rate limit promo ongoing; free tier access still active |
Frequently Asked Questions
How much does OpenAI Codex cost per month?
Codex is included in ChatGPT Plus ($20/month), ChatGPT Pro ($200/month), ChatGPT Business ($25/user/month billed annually, or $30 month-to-month), and ChatGPT Enterprise (custom pricing). The Free tier currently includes limited Codex access as a temporary promotion. API access is billed per token with no monthly subscription required.
What is the Codex usage limit on Plus?
ChatGPT Plus users get 45-225 local agent messages and 10-60 cloud tasks per 5-hour rolling window. The range is wide because task complexity determines actual consumption: a simple 1-2 file edit uses 1-3 messages, a medium 3-5 file task uses 5-15, and a complex 10+ file refactor can use 20-50 messages. Heavy users frequently hit limits mid-window.
What is the 5-hour rolling window in Codex?
Codex uses rolling 5-hour windows rather than monthly limits. Each window resets 5 hours after your first usage in that window, not at a fixed clock time. When you exhaust a window, you wait for reset or purchase credits to continue. This design means a multi-hour coding sprint can drain your entire allowance before a reset.
Is there a free version of Codex?
Yes, but it is a temporary promotional offer. The Free ChatGPT plan currently includes limited Codex access with no fixed end date announced. OpenAI has indicated this may change. For consistent access, Plus ($20/month) or a subscription tier is more reliable.
How much does the Codex API cost?
API pricing varies by model: codex-mini-latest costs $1.50/million input tokens and $6.00/million output tokens. gpt-5.2-codex costs $1.25 input and $10.00 output. gpt-5.3-codex is approximately $1.75 input and $14.00 output. gpt-5.1-codex-mini is the cheapest at $0.25 input and $2.00 output. A typical coding session costs $0.50-$2.00 at API rates.
When is the Codex API cheaper than a Plus subscription?
If you code fewer than 10-15 sessions per month, the API is cheaper than Plus ($20/month). At $0.50-$2.00 per session, you break even at roughly 10-40 sessions depending on session length. The API also provides access to all models without subscription tier restrictions, though you lose cloud features like Codex.com integration and GitHub sync.
Do Codex credits expire?
Yes. Codex credits (pay-as-you-go add-ons purchased when you exceed plan limits) expire 12 months from purchase. Your subscription's included usage is consumed first; credits are drawn on only after that is exhausted. There is no rollover from month to month for the included usage.
What does ChatGPT Pro include for Codex?
ChatGPT Pro at $200/month includes 300-1,500 local agent messages per 5-hour window and 50-400 cloud tasks per window (approximately 6x Plus limits). Pro users also get priority processing, access to GPT-5.3-Codex-Spark preview, and higher code review limits (100-250 per week vs 10-25 for Plus).
How does Codex Business pricing work?
ChatGPT Business costs $25 per user per month billed annually, or $30 per user per month-to-month. Usage limits are the same as Plus (45-225 local messages, 10-60 cloud tasks per 5-hour window). Business adds larger VMs for cloud tasks, admin controls, SAML SSO, MFA, and a guarantee that your data is not used for model training.
How does Codex Enterprise pricing work?
Enterprise pricing is custom and requires contacting OpenAI sales. Instead of fixed rolling-window limits, Enterprise uses a credits pool that the organization allocates. It includes priority processing, SCIM, EKM, RBAC, audit logs, and dedicated support. This tier targets large teams where the $25-$30/user Business plan becomes cost-prohibitive.
Building coding agents? Morph applies edits at 10,500 tokens/sec.
If you are building an AI coding agent or editor on top of Codex (or any model), Morph's apply model handles the code edit application step at a fraction of the inference cost. Most edit sessions cost under $0.01.