Codex pricing — how much does pay-as-you-go cost when the plan table does not say?

Search results for Codex pricing almost all lead to plan comparison tables. Often, though, the numbers you actually need are missing. That is because there are two pricing models, and plan tables explain only one of them. This article covers the other one—what token-based pay-as-you-go actually costs—with real prices, so you can decide which option is cheaper with a simple division.

The two are not combined. You can keep your plan and use a key.

Pay-as-you-go rates and the actual cost per task

Rates are read directly from the catalog, not entered by hand, so this text updates when prices change. GPT-5.6 Sol costs $2.00 / $12.00 per 1M tokens.

Rates alone are hard to visualize, so here is the cost per task. One agentic step using 25,000 input and 1,200 output tokens (enough to read code and make one change) costs about $0.0644 — or about 16 steps per $1. A task that takes 20 steps costs about $1.288.

# 내 사용 패턴에서 한 작업이 얼마인지 직접 재는 방법.
# 응답의 usage에 실제 토큰 수가 들어옵니다 — 추정하지 말고 재세요.
curl -sS https://api.kunavo.com/v1/responses \
  -H "Authorization: Bearer sk-kn-..." \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5-6-sol","input":"이 저장소의 테스트를 나열해줘"}'

The response’s usage contains the actual token count, so you can use it to determine the cost per task for your usage pattern. Measure it instead of estimating.

Plus or Pro — what to check first

Before comparing subscriptions, it is better to check whether you need a subscription at all. The calculation is one line.

Monthly fee ÷ cost per task = break-even number of tasks. Since one task costs about $1.288, the break-even point for the $20 plan is about 16 tasks. If you often use fewer than that in a month, neither subscription is the right choice.

Plan prices and limits change frequently, so we do not copy them into this page. As on the English Codex pricing page, we provide the calculation rather than figures that can change.

You’ve reached the limit but need to finish today

According to OpenAI Help, when you hit a limit the available options are to add credits, apply an available reset, upgrade, or wait until the reset time. Eligible Plus and Pro users can buy credits without changing plans. Upgrading to an available higher plan applies it immediately and restarts the billing cycle (Pro tier help). Because an upgrade starts a new billing cycle, if you only need to continue for that day, you can move just that task to a key without canceling your subscription.

# Codex CLI를 종량제 엔드포인트로 돌리는 두 줄.
# Codex가 받아들이는 건 Responses API뿐이고, chat completions로는 동작하지 않습니다.
export OPENAI_BASE_URL=https://api.kunavo.com/v1
export OPENAI_API_KEY=sk-kn-...

codex

There is one important constraint: the Codex CLI accepts only the Responses API for custom providers, so it will not work with a gateway that provides only chat completions. When choosing an alternative, check support for POST /v1/responses before looking at the price. For key creation and setup steps, see Get a Codex API key; for the English instructions, see Codex CLI API key setup.

Two ways to reduce costs

First, use different models for different tasks. There is no reason to use the top-tier model for supporting tasks such as summarizing, classifying, and organizing. GPT-5.6 Terra costs $0.70 / $4.20 per 1M tokens.

Second, use prompt caching. It greatly reduces input charges when your usage pattern sends the same context (repository description, coding rules) repeatedly. If you are deciding between Codex and Claude Code, Codex vs. Claude Code compares their billing models, and Claude Code pricing covers the cost of Claude Code.

Frequently asked questions