Here’s the thing that trips up almost everyone researching this: there is no such product as “a Codex subscription.” Codex isn’t sold on its own — it rides on top of your ChatGPT plan, its billing model changed fundamentally partway through 2026, and the actual cost you’ll pay depends on a combination of subscription tier, token consumption, and which model you’re running. None of that is obvious from the outside, so here’s the current picture, verified against OpenAI’s own pricing pages.
The Short Answer
Codex costs $0 to $200 per month for individuals, depending entirely on which ChatGPT plan you’re on — Free, Go ($8), Plus ($20), Pro 5x ($100), or Pro 20x ($200) — plus any additional credits you burn through beyond your plan’s included usage. Business runs $25/user/month (roughly $20/user billed annually, 2-seat minimum), with Enterprise on custom pricing. Since April 2, 2026, OpenAI retired flat per-message pricing entirely and moved Codex to token-based credit billing, meaning “one coding task” no longer maps to a predictable fixed cost the way it briefly did before that change.
Current Plan Pricing
| Plan | Price | What You Get |
|---|---|---|
| Free | $0/month | Trial-level access, limited Codex messages |
| Go | $8/month | Lightweight Codex access, still limited |
| Plus | $20/month | Full Codex across web, CLI, IDE extension, and iOS; cloud code review and Slack integration |
| Pro 5x | $100/month | 5x Plus’s rate limits |
| Pro 20x | $200/month | 20x Plus’s rate limits |
| Business | $25/user/month | Workspace-level access, SSO, no-training data guarantee (2-seat minimum) |
| Enterprise | Custom | Flexible, credit-based scaling |
Most professional developers actually using Codex regularly for real work report landing between $100 and $200 per month once subscription cost and credit overages are combined — light, occasional users can stay close to the $20 Plus tier, but genuinely heavy agentic coding sessions burn through included credits fast enough to push serious users toward Pro.
Why Billing Changed on April 2, 2026
This date matters because it changed the entire cost calculus for anyone who’d budgeted around the old system. Before this change, Codex billed per message — a flat, predictable cost per interaction. After it, OpenAI switched to token-based credit billing across Plus, Pro, Business, and Enterprise, meaning cost now depends on actual token consumption per task rather than a fixed per-message rate. This is a meaningfully less predictable system for budgeting, but it more accurately reflects the real computational cost of a task — a quick one-line fix and a full repository refactor were always going to cost fundamentally different amounts of compute, even though the old per-message system charged them identically.
Understanding the Three Token Types That Actually Drive Cost
Current per-1M-token credit rates vary by model — GPT-5.5 runs roughly 125 credits for input and 750 for output, GPT-5.4 runs about half that, and GPT-5.4 mini considerably less still. But the more useful thing to understand isn’t the specific numbers (which shift with each model release) — it’s the three categories driving your actual bill:
Input tokens cover everything sent to the model: your prompt, the system prompt, and any code or context loaded into the session. For agentic coding tasks reading large files or extended conversation history, input tokens frequently dominate total cost.
Cached input tokens are input tokens that hit OpenAI’s prompt cache, billed at a steep discount (historically around 75% off standard input rates) — repeated context across a session benefits meaningfully from this if your workflow reuses the same files or instructions repeatedly.
Output tokens are what the model actually generates — code, explanations, edits — typically billed at a notably higher rate than input tokens, since generation is more computationally expensive than reading.
The 5-Hour Rolling Window, Not a Monthly Allowance
This trips up a lot of new users: Codex usage limits reset on a rolling 5-hour window per seat, not a monthly quota the way most SaaS subscriptions work. Published usage ranges (commonly cited as 45-225 messages within a window) span a wide gap specifically because task complexity drives real consumption far more than a simple message count does — a handful of complex, repository-wide refactoring tasks can drain an entire Plus-tier window in as few as 2-4 interactions, while simple, targeted fixes stretch much further.
Subscription vs. API: Which Actually Costs Less for You
If your Codex usage is genuinely light — fewer than roughly 10-15 sessions per month — paying per-token through the API directly (commonly working out to $0.50-$2.00 per session for typical tasks) is often cheaper than committing to a $20/month Plus subscription you’re not fully using. The trade-off: API access skips the ChatGPT cloud conveniences — no automatic GitHub code review integration, no Slack integration — and newer models sometimes reach API users later than they reach subscription users. For unattended automation specifically (a scheduled CI pipeline, a background agent with no human in the loop), the API is the only realistic path regardless of volume, since it doesn’t depend on an interactive ChatGPT session at all.
How Codex’s Pricing Compares to Other AI Coding Tools
This is worth putting in context rather than evaluating in isolation. At the entry level, Codex via ChatGPT Plus ($20/month) sits right alongside Cursor Pro (also $20/month) and above GitHub Copilot Pro ($10/month) — genuinely comparable pricing at the entry tier, with the real differentiator being interface and autonomy rather than cost. Codex runs agentic, sandboxed tasks — including fully cloud-based tasks that open pull requests with no local editor involved at all — while Cursor and Copilot are fundamentally IDE-embedded tools you interact with directly inside your editor. If you’re evaluating Codex specifically against Cursor’s approach to deep codebase understanding, our Cursor comparisons cover that architectural difference in more depth than pricing alone can capture.
What Changed With the Model Lineup
Codex now runs on the GPT-5 series — GPT-5.5, GPT-5.4, and GPT-5.4 mini — a genuine shift away from the earlier codex-1 model built on the o3 family. Notably, OpenAI states GPT-5.5 uses significantly fewer tokens to achieve results comparable to GPT-5.4, meaning its higher per-token rate doesn’t necessarily translate to a higher effective cost per completed task — worth testing directly on your own workload rather than assuming the higher-priced model is automatically the more expensive choice in practice.
Frequently Asked Questions
Can I buy Codex as a standalone product without a ChatGPT subscription?
Not for interactive use — Codex capabilities are bundled into ChatGPT’s plan tiers rather than sold separately. Developers integrating Codex into automated pipelines or products can access it purely through the OpenAI API using a key, billed at standard per-token rates, without needing a ChatGPT subscription at all.
Why did my Codex costs become less predictable in 2026?
OpenAI retired flat per-message pricing on April 2, 2026, in favor of token-based credit billing. This ties cost more accurately to the actual computational work a task requires, but makes budgeting less predictable than the old fixed-rate system.
Is the API cheaper than a ChatGPT Plus subscription for Codex access?
It depends on your usage volume. Below roughly 10-15 sessions per month, per-token API billing is often cheaper than a $20/month Plus subscription. Above that, a subscription plan typically becomes the more cost-effective option.
How much does Codex typically cost a professional developer per month?
Most reports place real-world costs for active professional use between $100 and $200 per month once subscription fees and credit consumption are combined, though light users can stay closer to the $20 Plus tier.
Does Codex reset usage monthly like most subscriptions?
No — Codex enforces usage limits through a rolling 5-hour window per seat rather than a monthly quota, meaning your available capacity refreshes continuously rather than resetting on a billing-cycle date.
Pricing in this specific product category has shifted meaningfully more than once within 2026 alone, so verifying current rates directly against OpenAI’s own pricing page before budgeting is worth the extra minute — treating any single comparison article’s numbers, including this one, as a snapshot rather than a permanent rate card.
