Direct answer
Codex credits = tokens ÷ 1M × the model's credit rate
Since April 2, 2026, Codex charges credits per million tokens, with separate rates for input, cached input and output. On GPT-6.1 Sol that is 50, 2.5 and 250 credits per 1M tokens. A session with 300,000 uncached input, 2,100,000 cached input and 60,000 output tokens therefore costs 35.25 credits. Rates below were checked against OpenAI's Codex pricing page on November 8, 2026.
Codex credits for a session, a day and a month
Pick a model, enter one session's tokens, and set how often you run sessions like it. Rates are OpenAI's Codex credit rate card, checked November 8, 2026.
Credits per month
2,961
141 per day × 21 active days
Credits per session
35.25
GPT-6.1 Sol, Standard speed
Credits per day
141
4 sessions per active day
Uncached input
15 cr
300,000 tokens × 50 / 1M
Cached input
5.25 cr
2,100,000 tokens × 2.5 / 1M
Output (includes reasoning)
15 cr
60,000 tokens × 250 / 1M
API-equivalent dollars per month
$118.44
$1.41 per session. Credits ÷ 25, the standard API price for the same tokens. Not the price you pay for credits.
Paste token totals (optional)
Paste token lines from Codex's /status or /usage, or from the whoburnedmore local dashboard. The reader looks for numbers next to the words input, cached (or cache read) and output, and understands 45K or 1.2M. It skips cache-write and separate reasoning lines. The text stays in this page.
Plan context: local messages per five hours
OpenAI's published estimates for included plan usage, checked November 8, 2026. Low end means large, long-context tasks. These are not a credit conversion. Weekly limits may also apply, and cloud tasks may use more than local messages.
| Model | Plus | Business | Pro |
|---|---|---|---|
| GPT-6 Astra | 5–45 | 5–45 | no 5-hour limit |
| GPT-6.1 Sol (selected) | 15–160 | 15–160 | no 5-hour limit |
| GPT-6 Sol | 15–150 | 15–150 | no 5-hour limit |
| GPT-6.1 Terra | 35–285 | 35–285 | no 5-hour limit |
| GPT-6 Luna | 350–3,000 | 350–3,000 | no 5-hour limit |
- Enterprise and Edu, flexible pricing: No fixed rate limits.
- Enterprise and Edu, without flexible pricing: Same per-seat usage limits as Plus for most features.
- Free and Go: Lightweight access on GPT-6 Luna; no published numbers.
- API key: No plan limit or credits; billed at standard API prices.
Same tokens, every model
Compare models side by side
The same work on GPT-6 Luna would cost 2.03 credits per session, against 35.25 on GPT-6.1 Sol. All rows use Standard speed and the same token counts. A real session on another model may use a different number of tokens, so read this as same-work pricing, not a forecast.
| Model | Credits / session | Credits / month | vs GPT-6.1 Sol |
|---|---|---|---|
| GPT-6 Astra | 202.5 | 17,010 | 5.74× |
| GPT-6.1 Solselected | 35.25 | 2,961 | 1× |
| GPT-6 Sol | 40.5 | 3,402 | 1.15× |
| GPT-6.1 Terra | 17.63 | 1,481 | 0.5× |
| GPT-6 Luna | 2.03 | 170.1 | 0.06× |
| GPT-5.6 Sol | 81 | 6,804 | 2.3× |
| GPT-5.6 Terra | 43.5 | 3,654 | 1.23× |
| GPT-5.6 Luna | 4.35 | 365.4 | 0.12× |
Transparent math
How Codex credits are calculated
Each of the three token classes is billed at its own per-million rate, then the speed multiplier applies. There is no separate cache-write charge. Reasoning tokens are part of output, so they are not added a second time.
Since April 2, 2026
How Codex credits work since April 2, 2026
Tokens replaced per-message averages
The earlier rate card quoted average credits per message or cloud task. Now Plus, Pro, Business and new Enterprise plans are priced on the tokens a session actually moves. A small subset of Enterprise customers stays on the legacy rate card until OpenAI migrates them.
Included allowance first, credits after
On Plus and Pro your plan allowance is used first. Purchased credits cover usage beyond it and are valid for 12 months. Business, Edu and Enterprise workspaces on credit-based or flexible pricing buy workspace credits. API-key users never use credits; they pay standard API prices.
Modifiers on top of the rate card
Fast mode uses 2× credits and Ultrafast 6× (GPT-6 Astra and GPT-6.1 Sol). Desktop voice costs 1.25 credits per minute on Business, Edu and Enterprise credit billing, and when Plus and Pro users spend additional credits. Image generation has its own rates and is not modelled here.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 | 25 | 1,250 |
| GPT-6.1 Sol | 50 | 2.5 | 250 |
| GPT-6 Sol | 50 | 5 | 250 |
| GPT-6.1 Terra | 25 | 1.25 | 125 |
| GPT-6 Luna | 2.5 | 0.25 | 12.5 |
| GPT-5.6 Sol | 100 | 10 | 500 |
| GPT-5.6 Terra | 50 | 5 | 300 |
| GPT-5.6 Luna | 5 | 0.5 | 30 |
For who can buy credits, where, and how the meter compares with the old allowance, read the Codex credit billing guide. For the five-hour and weekly limits that apply before credits, read Codex usage limits.
Example dated November 8, 2026
Worked scenario: a long refactoring session on GPT-6.1 Sol
An illustrative session that keeps a large repository in context. Most input is a cached prefix re-sent each turn. The developer runs 4 sessions like it per day on 21 workdays. These are the calculator's default inputs.
| Token class | Tokens | Credits / 1M | Credits |
|---|---|---|---|
| Uncached input | 300,000 | 50 | 15 |
| Cached input | 2,100,000 | 2.5 | 5.25 |
| Output (incl. reasoning) | 60,000 | 250 | 15 |
| Per session | 2,460,000 | — | 35.25 |
Per day (4 sessions)
141 credits
Per month (21 days)
2,961 credits
API-equivalent per month
$118.44
The same work on other models
- GPT-6 Astra: 202.5 credits per session, 17,010 per month
- GPT-6.1 Sol: 35.25 credits per session, 2,961 per month
- GPT-6.1 Terra: 17.63 credits per session, 1,481 per month
- GPT-6 Luna: 2.03 credits per session, 170.1 per month
The cheapest model is not automatically the cheapest result: a weaker model may need more turns, and more tokens, to finish the same refactor.
Inputs
Where to get your token counts
Inside the Codex CLI, /status shows the current session's token usage and /usage shows account token activity. For history by day and model, run the whoburnedmore CLI command npx whoburnedmore --local. It reads Codex's local session logs, splits input, cached input and output, and writes an offline dashboard without uploading anything. See how to check Codex CLI usage for every option.
Limits of the estimate
What this calculator cannot see
It does not read your credit balance, remaining allowance or purchase price; the ChatGPT usage dashboard at chatgpt.com/codex/settings/usage is authoritative for those. It also misses tokens that never reached your local logs: cloud tasks, code reviews, other devices and shared ChatGPT features. Treat the result as a same-work estimate, then reconcile it with the dashboard.
Frequently asked questions
How many Codex credits does a task use?
It depends on the tokens the task moves and the model. OpenAI's own guide figure is that a typical GPT-5.6 Sol task may use 2–15 credits. Large repositories, long sessions and high reasoning effort push a task above that, because every turn re-sends context and reasoning is billed as output. Enter your real token counts above rather than relying on an average.
How much is one Codex credit in dollars?
OpenAI does not publish a dollar price per credit. Its pricing page says credit purchase prices depend on your plan or agreement, so check the purchase screen in ChatGPT Settings → Usage. This calculator's dollar figure is API-equivalent only: every listed text model's credit rate is exactly 25 times its standard API price, so credits ÷ 25 is what the same tokens would cost on an API key at standard short-context rates.
Are cached tokens counted twice?
They should not be. Codex's session logs report cached tokens as part of the input total. If your numbers come from such a source, tick the box that says your input already includes cached tokens, and the calculator charges only the uncached remainder at the full input rate. The whoburnedmore dashboard already splits uncached input from cache reads, so leave the box unticked for its numbers. Codex credits have no cache-write charge.
Do credits count against my five-hour limit?
No. On Plus and Pro, the included plan allowance is used first, with its five-hour and weekly limits. Purchased credits only cover usage beyond it. The five-hour table shows OpenAI's local-message ranges for that included usage. It is not a credit conversion, and the calculator does not read your remaining allowance or credit balance.
What do Fast and Ultrafast modes do to the credit count?
For purchased credits and Enterprise pay-as-you-go usage, Fast mode uses 2× the Standard rate and Ultrafast uses 6×. Ultrafast is listed for GPT-6 Astra and GPT-6.1 Sol. Against included plan usage the multipliers are higher: 2.5× for Fast and 8× for Ultrafast. Pick the speed in the calculator to apply the credit multiplier.
Why does my credit balance fall faster than this estimate?
Local token totals only cover Codex sessions on the machine they came from. Cloud tasks, GitHub code reviews, other devices, desktop voice and other ChatGPT features that share the allowance all draw from the same plan without showing up in local logs. Image generation has its own rates. Compare the calculator with the usage dashboard over the same days before you trust a monthly forecast.
Sources
- OpenAI Codex pricing, limits and credit rate card
Credits per 1M input, cached input and output tokens by model; no separate cache-write charge; Fast 2× and Ultrafast 6× on purchased credits; local messages per five hours by plan. Checked November 8, 2026.
- OpenAI API pricing
Standard short-context API prices. Each listed Codex credit rate is exactly 25 times the matching API dollar price, which is the basis of the API-equivalent view. Checked November 8, 2026.
- OpenAI Help Center: Using credits for flexible usage
How Plus and Pro users buy and spend credits after the included allowance. Checked November 8, 2026.