Codex pricing is not a single number
Codex is OpenAI's coding agent, bundled into ChatGPT plans rather than sold standalone. Money moves four ways: subscription usage, flexible credits, API-key billing at per-token rates, and a per-token route that keeps the Codex harness but swaps the billed model.
On 2026-04-02 OpenAI retired per-message pricing and moved Codex to token-based credit billing, so "one task" no longer maps to a fixed cost. This guide separates the four meters, converts credits to dollars, and prices real workloads, all verified against first-party sources on 2026-08-22.
Current Codex pricing at a glance (last verified 2026-08-22)
Current individual and team prices, verified against OpenAI's Codex pricing page:
| Plan | Price | What it is |
|---|---|---|
| Free | $0/mo | Trial-grade quick tasks |
| Go | $8/mo | Lightweight coding |
| Plus | $20/mo | Full GPT-5.6 + cloud features |
| Pro | From $100/mo | 5x or 20x Plus limits |
| Business | $20/user/mo | Usage-based team seats |
| Enterprise & Edu | Contact sales | Custom, flexible pricing |
Assumptions: Codex ships inside ChatGPT plans, not standalone. Business is $25/user/mo billed monthly (2+ users); Pro is $100/mo (5x) or $200/mo (20x).
Free is trial-grade quick tasks, Go covers lightweight work, and Plus is the first tier for real weekly sessions with the full GPT-5.6 family and cloud integrations. Pro adds two rate-limit tiers, 5x or 20x Plus usage, plus the Codex-Spark research preview. Business seats are usage-based; Enterprise and Edu are contact-sales with flexible pricing and, when enabled, no fixed rate limits.
Is Codex free?
Yes and no. The Free tier ($0) explores quick coding tasks but is trial-grade — it omits cloud features (GitHub code review, Slack) and hits limits fast. Sustained work needs a paid plan, and the two per-token routes (an OpenAI API key or the Fast alternative) are metered, not free. Once you pass included usage, you buy credits or switch routes.
The four meters that decide what Codex costs
Codex cost is never one blended bill. Four meters run in parallel, and mixing them is why people misjudge what they'll pay.
| Meter | What you pay | Billed as | Best for |
|---|---|---|---|
| Subscription usage | Flat monthly fee | Included credits | Steady daily work |
| Flexible credits | Top-up over quota | Credits | Bursts past limits |
| API-key billing | Per token | Dollars | CI, automation |
| Fast per-token | Per token | Prepaid credits | Swapping the model |
Assumptions: Meters run in parallel, not blended. API-key and Fast routes drop cloud features.
Meter 1 — Subscription usage. A flat monthly fee buys a pool of included usage, metered in credits and refreshed on a rolling five-hour window — what most people mean by "codex subscription."
Meter 2 — Flexible credits. When included usage runs out, you buy flexible credits — personal for Plus/Pro, shared for Business/Edu/Enterprise workspaces. Credits are priced per 1M input, cached-input, and output tokens, varying by model. One ChatGPT credit ≈ $0.04 (each credit rate ÷ its API dollar rate; Sol 100 credits/1M ÷ $4.00 = $0.04). Fast mode burns credits faster where supported (2.5x Standard for GPT-5.5, 2x for GPT-5.4).
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-5.6 Sol | 100 | 10 | 500 |
| GPT-5.6 Terra | 50 | 5 | 300 |
| GPT-5.6 Luna | 5 | 0.5 | 30 |
| GPT-5.5 | 125 | 12.5 | 750 |
| GPT-5.4 | 62.5 | 6.25 | 375 |
| GPT-5.4 mini | 18.75 | 1.875 | 113 |
Assumptions: Credits per 1M tokens. One ChatGPT credit ≈ $0.04.
Meter 3 — API-key billing. Sign into Codex with an OpenAI API key and pay standard per-token rates directly. This route drops cloud features, so you pay only for tokens consumed.
| API model | Input | Cached input | Output |
|---|---|---|---|
| gpt-5.3-codex | $1.75 | $0.175 | $14.00 |
| gpt-5.3-codex (Fast) | $3.50 | $0.35 | $28.00 |
| gpt-5.6-sol | $4.00 | $0.40 | $20.00 |
| gpt-5.6-terra | $2.00 | $0.20 | $12.00 |
| gpt-5.6-luna | $0.20 | $0.02 | $1.20 |
Assumptions: USD per 1M tokens, Standard short-context pricing.
Meter 4 — Fast per-token. Keep the Codex harness but route model calls to a separately billed Fast Inference catalog model, a labeled second path rather than more OpenAI quota (full setup below).
Cost per successful task, by workload
A task is a bundle of input, cached-input, and output tokens: multiply each by its credit rate, sum to credits, then multiply by ≈$0.04 for dollars.
| Workload scenario | Tokens (in → out) | Credits | ≈ Cost | Cheapest route |
|---|---|---|---|---|
| Light edit | 20K → 2K | 3 | $0.12 | Plus |
| Feature session | 200K → 20K | 30 | $1.20 | Plus |
| Heavy refactor | 1M → 100K | 150 | $6.00 | Pro 5x |
| CI batch | 5M → 500K | 750 | $30.00 | API key |
Assumptions: GPT-5.6 Sol API rates ($4.00 input, $20.00 output per 1M), 1 credit ≈ $0.04; token bundles are representative.
OpenAI's own estimate for heavy Codex use is roughly $100–$200 per developer per month, with variance from model choice, parallel instances, automations, and Fast mode.
The Fast Inference alternative route
If none of the four options fit, there is a reversible fifth: keep the Codex harness and route to a metered model served by Fast Inference, billed per token from a separate balance. A new Fast account starts with $26 in free credit. Install the fast CLI, log in, and turn the route on; the default model is glm-5.2, and --model selects any other catalog model.
# 1. Install the fast CLI and sign in (selects your first project).
npm install -g @inference/fast
fast login
# 2. Route the Codex harness through a metered Fast Inference model.
# Default model is glm-5.2; the harness is unchanged.
fast codex on
# ...or select any other callable catalog model instead of the default.
fast codex on --model '<catalog-id>'
# 3. Confirm the route is active before sending real traffic.
fast codex status
# 4. Restore the original ~/.codex/config.toml when you are done.
fast codex offfast codex off restores your original ~/.codex/config.toml. Usage is billed per token with credits drawn first from your prepaid balance — check per-token rates and pick a catalog model before routing traffic. This reversible path preserves the harness while routing model calls to a separately billed model; it does not increase native OpenAI quotas or accelerate an unchanged proprietary model.
Which route is cheapest for you?
Lightest to heaviest: Free/Go for casual work, Plus ($20/mo) for a few weekly sessions with cloud features, Pro 5x ($100/mo) for daily use that hits the five-hour limits, Pro 20x ($200/mo) for all-day or parallel use, an API key for CI and automation billed strictly per token, and Fast per-token to keep the harness while choosing a separately billed model.
To stretch the plan you already have, check and reduce Codex usage instead of switching routes, or compare Codex and Claude Code if you are still choosing a harness.