Home / Guides / How Much Does Codex Cost
How Much Does Codex Cost? A 2026 Pricing Breakdown
On paper, OpenAI Codex pricing is simple: $20 or $200 a month through ChatGPT. But the April 2026 switch to token-based metering changed what those dollars actually buy you. Here's the real breakdown, plus the pay-per-token API option and how to track what you spend.
Last updated June 2026 · By Soren Starck
The Short Answer
Codex itself does not have a standalone subscription. You get it bundled with an OpenAI ChatGPT plan, or you pay per token through the OpenAI API. There are three common paths:
- ChatGPT Plus ($20/month) – the cheapest way to use Codex. Smaller token allowance, fine for short bursts.
- ChatGPT Pro ($200/month) – roughly 20x the Plus token allowance, plus access to OpenAI's top reasoning models. Built for all-day Codex CLI use.
- OpenAI API (pay-per-token) – no subscription, you pay only for tokens consumed. Predictable per-task, unpredictable per-month.
One thing to keep straight throughout this guide: the prices above are what you pay OpenAI for Codex access. That is completely separate from SessionWatcher, the $6.99 one-time macOS app that tracks your Codex usage. We'll cover SessionWatcher pricing at the end and keep it clearly labeled.
Codex Subscription Tiers (What You Pay OpenAI)
Codex access comes through ChatGPT subscriptions. The price is fixed; what varies is how many tokens you can burn inside the rolling 5-hour window before you get rate limited.
| Plan | Monthly Cost | Codex Allowance (per 5h) | Best For |
|---|---|---|---|
| ChatGPT Plus | $20/mo | ~120K–800K tokens | Short bursts, general ChatGPT + light Codex |
| ChatGPT Pro (5x) | $100/mo | ~0.7M–4M tokens | Daily Codex use plus GPT-5.5 for general work |
| ChatGPT Pro (20x) | $200/mo | ~2.5M–16M tokens | All-day Codex CLI, parallel agents, top reasoning |
The token allowances above are ranges, not hard published numbers – OpenAI does not publish an exact token cap per tier, and the figures fluctuate with promo boosts (the $100 Pro tier was promo-boosted toward 10x Plus through May 2026). Treat them as order-of-magnitude guidance, not contractual limits. Each tier is also subject to a separate weekly cap on top of the 5-hour window.
How the April 2026 Token Rate Card Changed the Math
Before April 2026, Codex limits were framed loosely as a number of messages. Since then, OpenAI meters Codex by tokens consumed inside a rolling 5-hour window, plus a weekly token cap. This matters because your subscription dollar cost did not change, but what those dollars buy did.
The key insight from OpenAI's own rate-card documentation: agentic Codex work (the CLI editing files, running tools, reading large context) burns roughly 3 to 10 times the tokens of a plain chat user. A single autonomous agent run can consume tens of thousands of tokens, and a heavy CLI session pushes 150K–400K tokens per hour. That is why a Plus plan that feels generous for chat can lock you out in well under an hour of intensive Codex agent work.
In other words: under the token model, your effective cost per useful Codex hour depends entirely on your workflow. Light chat-style prompting stretches the Plus plan a long way. Heavy agentic coding can exhaust even Pro allowances and push you toward the API. For more on how the window itself behaves, see our Codex rate limits guide.
The Pay-Per-Token API Option
Instead of a subscription, you can run Codex against the OpenAI API and pay only for what you use. No monthly fee, no rate-limit window – just metered tokens. Here are the published API rates for the relevant models (USD per million tokens):
| Model | Input / M | Output / M | Cached Input / M |
|---|---|---|---|
| GPT-5.3 Codex | $2.50 | $15.00 | $1.25 |
| GPT-5 | $1.25 | $10.00 | $0.625 |
| GPT-5 mini | $0.25 | $2.00 | $0.125 |
Output tokens cost roughly 5–6x more than input, and cached input is about half price. Codex sessions are output-heavy when generating code, which is why API bills add up faster than the input rate alone suggests.
Worked Examples
Let's put real numbers on it. These use GPT-5.3 Codex API rates ($2.50 input / $15 output per million), assuming a typical agentic split of about 80% input / 20% output before caching.
Example 1: A single feature task (~200K tokens)
160K input + 40K output. That's roughly 160K × $2.50/M = $0.40 input, plus 40K × $15/M = $0.60 output ≈ $1.00 per task on the API. Caching frequently-read context can cut the input portion meaningfully.
Example 2: A heavy 1-hour agentic session (~300K tokens)
240K input + 60K output ≈ $0.60 + $0.90 = roughly $1.50 per hour at the mid case, climbing toward $5–$10/hour on context-heavy fullstack work where the same large files are re-read. Over a 6-hour coding day that's anywhere from ~$9 to $50+ in API spend.
Example 3: Subscription vs API break-even
If your API spend regularly exceeds ~$20/month, ChatGPT Plus is cheaper (until you hit its token window). If it regularly exceeds ~$100–$200/month, a Pro subscription almost always wins on raw cost – provided your usage fits inside the rate-limit windows. The crossover point is entirely a function of how many tokens you burn, which is exactly why you need to measure it. Plug your own workflow into the free plan calculator to see which path is cheapest for you.
SessionWatcherStop guessing your Codex spend.
Track every token in real time.
Native macOS menu bar app. Track Claude and Codex usage, costs, and rate limits in real-time.
“Fast, simple, and does exactly what it should. Definitely worth it.”
@nicojerome on GitHub
macOS 14+. 7-day Bundle trial. No credit card.
The Hidden Cost: Rate-Limit Lockouts
On a subscription, the headline price is only half the story. The token-based metering means heavy Codex users hit the 5-hour window and get locked out mid-task, with no advance warning and no clear ETA. That lost time is a real cost that never shows up on an invoice.
The math is brutal. If your time is worth $75/hour and you lose just 2 hours/week to Codex lockouts:
2 hours/week × $75/hour = $150/week in lost productivity
$150/week × 4 weeks = $600/month in lost productivity
That $600 dwarfs the gap between the $20 Plus and $200 Pro plans. The way to avoid paying it is to see your token burn climbing before you hit the wall, so you can pace your work or wrap up the current task.
How to Track Your Actual Codex Cost
There are two practical ways to know what you're really spending:
1. The OpenAI usage dashboard: shows historical token usage and API spend, but with a delay. It won't tell you where you stand in your current 5-hour Codex window or warn you before a lockout.
2. SessionWatcher: SessionWatcher for Codex sits in your macOS menu bar and shows token usage, estimated dollar cost, your percentage of the rate limit, and the 5-hour window countdown – all in real time. You see the number tick up as you work, so you always know whether you're about to get locked out or burn an unexpected API bill.
SessionWatcher Pricing (Separate From OpenAI)
To be explicit: everything above is what you pay OpenAI for Codex. SessionWatcher is a separate, independent tool that simply tracks that usage. It is a one-time purchase, not a subscription:
| SessionWatcher | Price | What You Get |
|---|---|---|
| For Codex | $6.99 one-time | Real-time Codex usage, cost, and 5-hour window tracking |
| Bundle | $14.99 one-time | Codex plus the other supported coding tools |
| Pro | $49 one-time or $24/year | All tools plus advanced features |
All purchases come with a 30-day refund. At $6.99 once, SessionWatcher pays for itself the first time it helps you avoid a lockout or catch a runaway API bill – a fraction of a single month of any OpenAI plan.
SessionWatcherKnow your real Codex cost.
$6.99 once. Pays for itself day one.
Native macOS menu bar app. Track Claude and Codex usage, costs, and rate limits in real-time.
“Fast, simple, and does exactly what it should. Definitely worth it.”
@nicojerome on GitHub
macOS 14+. 7-day Bundle trial. No credit card.
Frequently Asked Questions
How much does Codex cost per month?
Codex is bundled with a ChatGPT subscription. ChatGPT Plus is $20/month with the smaller token allowance, ChatGPT Pro is $200/month for roughly 20x those limits, and there is a $100/month Pro (5x) tier in between. You can also skip subscriptions and pay per token through the OpenAI API.
How did the April 2026 token rate card change Codex cost?
OpenAI now meters Codex by tokens consumed in a rolling 5-hour window plus a weekly cap, instead of a fixed message count. Your subscription dollar cost is unchanged, but agentic Codex work burns 3–10x the tokens of chat, so heavy CLI users hit the limit much faster than the old message framing suggested.
What does Codex cost per token on the API?
GPT-5.3 Codex is roughly $2.50 per million input tokens and $15 per million output tokens, with cached input around $1.25. A heavy agentic session burning 150K–400K tokens/hour can run from about $1.50 to $10+ per hour depending on context size and caching.
How can I track my Codex costs?
SessionWatcher for Codex ($6.99 one-time for macOS) tracks token usage, estimated dollar cost, rate-limit percentage, and the 5-hour window in real time from your menu bar. The OpenAI dashboard shows historical spend but with a delay.