OpenAI Codex for Teachers: Plans, Credit Pricing and a Classroom Utility Walkthrough (September 2026)

An ICT teacher stays behind in the computer room after the bell. The terminal is still lit: "I want an offline grouping tool on the intranet so I stop drawing names from a hat every lesson." We remind him of a line schools rarely quote — OpenAI's own documentation puts Codex at roughly US$100 to US$200 per user per month in credits, on top of the ChatGPT subscription. He nods. "Right. So I buy Plus for myself, not Business for all hundred staff."

Codex is an agentic coding tool for people who read diffs and use git. It is not a lesson-planning platform and not a pupil learning environment. In Hong Kong schools it belongs on the desks of the few who will actually use it, with seat fees kept separate from credit spend.

What Codex is in 2026

Today's Codex has nothing to do with the 2021 model of the same name. It is OpenAI's agentic coding product line: a terminal CLI, IDE extensions for VS Code and others, web and desktop apps, an iOS app, automatic GitHub code review, and Slack and Linear integrations. Underneath runs the GPT-5.6 family (Sol, Terra, Luna) and GPT-6 Astra, released in September 2026.

In a school the people who can genuinely use it are ICT teachers, the IT coordinator who maintains internal systems, and subject teachers automating repetitive admin data work. OpenAI's teacher plan states plainly that it is not for students — that is product design, not an oversight.

Capabilities teachers actually touch

Run codex in a project directory and the local agent reads files, edits them and runs commands on your machine, asking for approval at each step if you want. Cloud tasks hand work to a remote environment that keeps running after you close the laptop, then pull results back locally. An AGENTS.md file in the repository root sets rules and conventions Codex reads at the start of every session. /review produces a prioritised findings list against uncommitted changes, a commit or a branch without modifying files. /permissions decides what Codex may do without asking — configure this first on a school machine. codex exec runs non-interactively for batch workflows. Skills and plugins package repeated instructions and connect GitHub, Gmail, Google Drive and more. Switch models with /model: Sol for the hardest reasoning, Terra as the everyday workhorse, Luna for fast, credit-light work.

Building an offline classroom utility

Scenario and prerequisites

You teach ICT in a secondary school and want a single HTML file for the intranet: paste a class list, generate random groups, run a lesson timer. No server, no internet, no pupil data collected. Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise. Free only covers tiny tasks; for real use, start at Plus (US$20 a month).

Install, permissions and AGENTS.md

On macOS or Linux:

curl -fsSL https://chatgpt.com/codex/install.sh | sh
mkdir class-toolkit && cd class-toolkit
codex

On first run choose "Sign in with ChatGPT". Then type /permissions and pick a mode that requires confirmation before editing files or running commands. Run /init to generate AGENTS.md and add:

This project builds single-file web utilities for Hong Kong secondary school teachers.

Rules:
- All output must be one index.html file with no external CDN and no network calls.
- Interface text in Traditional Chinese (Hong Kong usage), for example 分組, 計時, 重設.
- No analytics, no cookies, and no code that sends data out of the browser.
- The class list lives only in browser memory and is cleared on refresh.
- Comment the code in Chinese so another teacher can maintain it later.

Task, review and handover

Pick GPT-5.6 Terra with /model, then describe the goal:

Build index.html, an offline classroom tool with two tabs.

Tab one, "Random groups":
- A textarea, one student name or number per line
- Set either students per group or number of groups
- A "Group" button showing each group as a card, plus "Regroup"
- An "Export text" button copying results to the clipboard

Tab two, "Lesson timer":
- Countdown in minutes with start, pause and reset
- Background turns orange at one minute remaining, red and flashing three times at zero
- Type large enough to read from the back row on a projector

Use HTML, CSS and vanilla JavaScript in one file.

Read what Codex proposes. Look for fetch, XMLHttpRequest or an external <script src="http...">. Run /review and choose "Review uncommitted changes". Open index.html in a browser, unplug the network and try again. Pass the file to IT for the intranet. For the next utility:

codex exec "Following AGENTS.md, build seating.html: a computer-room seating planner with draggable student cards onto desk slots, printable."

Check /status every couple of weeks for model, permissions and remaining allowance.

School-built utilities rarely fail on functionality. They fail by quietly shipping a class list out of the browser — the three offline and no-tracking rules in AGENTS.md matter more than any prompt trick.

September 2026 pricing and credits

Codex comes with ChatGPT (US dollars, exclusive of tax): Free US$0; Go US$8/month; Plus US$20/month as the teacher entry point; Pro 5x / 20x at US$100 / US$200; Business standard seat US$20/user/month annual or US$25 monthly; Business Premium US$100/user/month annual; Enterprise / Edu custom credit pricing.

Credits per million tokens: GPT-6 Astra input 250 / output 1,250; GPT-5.6 Sol 100 / 500; Terra 50 / 300; Luna 5 / 30. A typical Sol task consumes 5 to 30 credits; Astra fast mode is 2.5x standard. One hundred Business standard seats at the annual rate is about US$24,000 a year (HK$187,000) — the wrong purchase when only two or three staff need Codex. ChatGPT for Teachers is free through June 2028, US K-12 only; Hong Kong does not qualify. GPT-5.4 retired in Codex on 31 August 2026.

How Codex sits beside alternatives and Edor.ai

Codex is terminal-first with weak classroom demo; Cursor at the same price projects better. Claude Code is similarly CLI-led; Base44 needs no coding. Codex wins on code; Edor.ai wins on school teaching, pupil gating, fixed annual fees and grant reporting — see school knowledge base, the HK$500,000 grant and the nine-tool overview.

Common misreadings

• The claim: Plus at US$20 a month means Codex costs US$240 a year all in. What actually happens: US$20 is the seat. OpenAI's documentation puts average credit spend at US$100 to US$200 per developer per month; heavy Sol or fast-mode use runs higher. Luna's rate is one twentieth of Sol's — routine utilities should run on Terra or Luna.

• The claim: Codex is so capable that Business for every teacher is the smart buy. What actually happens: One hundred standard seats at the annual rate is about US$24,000 a year, but most teachers need lesson planning and marking, not a CLI. In practice two or three competent users get Pro or Business seats; everyone else uses a school platform such as Edor.ai.

• The claim: Pupils learning Python should use Codex as their formal AI environment. What actually happens: Minimum age 13, no class management, no live teacher view of conversations. The teacher plan excludes pupils. Codex can be demonstrated in coding lessons; daily pupil AI belongs on a gated, audited school platform.

Three things to try on Monday

1. Open a Free account with your school email, run the grouping-tool flow in this article, and note credits consumed for one completed task via /status — that is your personal test of whether Plus is enough.
2. Save an AGENTS.md template with the three rules — no outbound data, no CDN, class list in memory only — in the panel shared drive for every future Codex project.
3. Confirm with IT that Codex installs on two or three teacher workstations only, not as Business for all staff, and put "US$100–200 average monthly credits" in the budget memo so mid-year overspend does not surprise the panel.