ChatGPT Edu offers several ways to work with AI, and each option can affect how credits are used. Faculty and staff can use this guide to choose among ChatGPT, ChatGPT Work, and Codex; select an appropriate model and reasoning effort; and respond when a usage limit is reached.
Key takeaway: Choose the lowest-cost experience and model that can do the job well. Reserve agentic experiences and higher reasoning efforts for work where they clearly improve the outcome.
Credit Usage
ChatGPT Edu provides broad access to AI tools. Credits support certain advanced models and agentic features. Think of credits as a shared institutional resource: use everyday options for routine work, and reserve higher-powered tools for tasks where their added capability makes a meaningful difference.
Credit use depends on the experience, model, reasoning effort, task complexity, and workspace configuration. Prompts, files, chat history, tool results, and generated outputs can all contribute to usage. Available models and limits may change, and Johns Hopkins University (JHU) may configure access or limits by user or group.
Choosing the Right Experience
A practical rule of thumb is to use ChatGPT for a conversation, Work for a deliverable, and Codex for software work.
| If you need to... | Start with | Why | Examples |
|---|---|---|---|
| Brainstorm, revise prose, explain a concept, draft an email, summarize a short document, or ask a quick question. | ChatGPT Instant | Fast, conversational, and appropriate for most everyday work. | Draft a concise, welcoming email to students explaining the deadline-extension process. Use a supportive tone and keep it under 200 words. |
| Compare options, analyze a complex policy, develop a careful lesson or assessment plan, or work through a difficult problem. | ChatGPT Thinking | Better suited to tasks where reasoning quality matters more than immediacy. | Using the attached course outcomes and assignment brief, identify alignment gaps and propose three revisions. Return a table with outcome, evidence, gap, and recommended change. |
| Research, analyze multiple files, or create a polished document, spreadsheet, presentation, report, or other deliverable. | ChatGPT Work | Designed for longer, multi-step knowledge-work tasks. | Review the attached program-review materials and create a two-page committee briefing. Include key findings, unresolved questions, and a recommendation section. Use only the attached files. |
| Write, debug, test, review, or modify software; work with a repository; or automate a technical workflow. | Codex | Designed specifically for software development and technical tasks. | Review this repository's grading-tool configuration. Identify the cause of the failed test, make the smallest safe fix, run the relevant test, and summarize the change. |
Choose a Model Based on Task Complexity
After choosing an experience, select the least resource-intensive model likely to succeed. Model choice and reasoning effort are separate settings: the model determines the underlying capability, while reasoning effort controls how much analysis the model applies to the task.
| Model | Best use | Credit-conscious guidance |
|---|---|---|
| Sol | Complex, ambiguous, high-value, or multi-step work, including difficult analysis, challenging debugging, sophisticated deliverables, or work requiring stronger judgment. | Reserve for tasks where Terra has not produced an adequate result or where the task clearly warrants added capability. |
| Terra | Most substantial Work and Codex tasks, including document creation, file analysis, lesson-material revision, typical coding, and troubleshooting. | Use as the everyday, credit-conscious starting point recommended by the Whiting School of Engineering (WSE). |
| Luna | Straightforward, well-scoped tasks; routine transformations; small fixes; or repeatable work. | Use when the quality requirements are clear and the task is bounded. Luna is a strong choice for routine or high-volume work when available. |
WSE credit-conscious recommendation: Use Terra for most Work and Codex tasks, Luna for bounded or repeatable work, and Sol for unusually complex, ambiguous, or high-value tasks. Start with Medium reasoning and increase it only when the work requires deeper analysis. Reserve Max and Ultra for exceptional cases.
Terra is a credit-conscious starting point for many Work and Codex tasks. OpenAI currently defaults the Power setting to Sol with Medium reasoning.
Choose Reasoning Effort Deliberately
Reasoning effort controls how much analysis a model applies to a task. Higher reasoning can help with difficult work, but it can also increase credit use. Start with Medium reasoning for most Work and Codex tasks, then increase it only when the task requires deeper analysis.
- Light or Low: Use for straightforward, well-defined work such as routine transformations, simple fixes, and tasks with clear instructions.
- Medium: Use as the starting point for most document creation, file analysis, coding, troubleshooting, and other substantial tasks.
- High or Extra High: Consider for work involving multiple constraints, careful comparison, difficult debugging, or complex analysis when Medium has not produced an adequate result.
- Max or Ultra: Reserve for exceptional cases involving unusually complex, ambiguous, or high-value work.
Before increasing the reasoning effort, improve the request by stating the goal, audience, source material, constraints, desired format, and definition of done. A complete, focused request can reduce rework and may save more credits than changing settings.
Practical Ways to Conserve Credits
- Start with a small test for ambiguous or credit-intensive tasks. Ask for an outline, plan, sample section, or proposed approach before requesting a full report, deck, site, or complex build. For a well-defined task, request the needed deliverable directly to avoid unnecessary turns.
- Provide a complete brief. Include the audience, purpose, source materials, constraints, tone, length, and desired output to reduce rework.
- Use one focused task. Avoid launching overlapping Work or Codex tasks for the same problem. Review the first result and request targeted revisions.
- Reuse relevant context. Keep related files and instructions in the same Project, but start a fresh chat when earlier conversation history is no longer needed.
- Request targeted revisions. Ask to revise the relevant sections rather than starting over when most of a draft is usable.
- Reserve agents for work that needs them. Work and Codex are most valuable when a task requires multiple steps, files, tools, or verification.
- Set a stopping point. Define the scope, such as “create a one-page draft” or “use only the attached files.”
View Your Remaining Usage
To view your remaining usage:
- Select your profile menu in the lower-left corner of ChatGPT.
- Review Usage remaining. The menu displays the percentage remaining for the current monthly limit.
If You Reach a Credit or Usage Limit
The JHU ChatGPT Edu workspace uses tiered group usage limits.
If you reach a workspace-configured usage limit, new credit-based Work or Codex tasks may pause. An active task may be allowed to finish. Chat and available Instant capabilities will remain accessible.
If this happens:
- Check your usage dashboard for information about your current usage or limit.
- Use Chat or an available Instant capability for work it can handle.
- Contact the Center for Media and Technology Solutions (CMTS) if additional access is needed.
Reset timing depends on the limit configured for your account.
Use AI Responsibly
Follow institutional requirements for data handling, privacy, accessibility, academic integrity, and records retention.
Do not enter protected health information (PHI) or patient health data into ChatGPT Edu. For work involving PHI, use a specifically approved JHU service that complies with the Health Insurance Portability and Accountability Act (HIPAA), and follow applicable Institutional Review Board (IRB), privacy, and other institutional requirements.
For additional guidance on using AI safely and appropriately at WSE, review Responsible Use of AI Tools at WSE.