UMD ChatGPT Edu

Best Practices & Usage Guidance

How to use ChatGPT modes effectively, and how to choose the right model, speed, and thinking level for Codex—so you get good results without burning unnecessary credits.

ChatGPT

Modes

Practical guidance for choosing the right mode for common campus work.

Fast response

Instant

The default model for rapid, everyday tasks where reasoning overhead isn't needed.

Best fit
Drafts, summaries, routine Q&A
Typical tasks
Short-form writing, quick lookups, reformatting

Real-world examples

  • Draft a follow-up email to a department committee
  • Summarize meeting notes before a faculty discussion
  • Answer a student's basic question about syllabus formatting
Balanced reasoning

Thinking Standard

Moderate reasoning for tasks that benefit from structured thinking without extended analysis.

Best fit
Analysis, planning, argument outlines
Typical tasks
Comparing options, evaluating tradeoffs, structured breakdowns

Real-world examples

  • Compare two course redesign approaches and recommend one
  • Outline a grant proposal section with supporting rationale
  • Identify common themes across a set of student feedback responses
Deeper analysis

Thinking Extended

Extended reasoning for complex problems that require careful, multi-step thinking.

Best fit
Complex analysis, nuanced arguments, multi-step logic
Typical tasks
Policy evaluation, decision support, methodological review

Real-world examples

  • Analyze a budget reallocation across competing departmental priorities
  • Evaluate a policy proposal with multiple stakeholder interests
  • Design a research study with methodological tradeoffs
Longer workflows

Pro Standard

Multi-step, agentic tasks where the model works through a sequence of actions to produce coordinated outputs.

Best fit
Extended workflows, multi-part deliverables
Typical tasks
Document sequences, iterative refinement, coordinated outputs

Real-world examples

  • Draft, review, and refine a department communication plan in one session
  • Generate a structured literature synthesis with citations from uploaded sources
  • Build a multi-section report from a collection of uploaded documents
Most comprehensive

Pro Extended

Maximum reasoning depth for high-stakes, precision-critical campus work where exhaustive analysis matters.

Best fit
High-stakes decisions, exhaustive synthesis
Typical tasks
Strategy memos, comprehensive review, rigorous analysis

Real-world examples

  • Write a research funding strategy memo with supporting evidence
  • Synthesize competing findings across multiple studies for a policy brief
  • Review and critique a draft policy document for gaps and inconsistencies

Codex

Session Calibration

Choose your starting settings intentionally — the wrong default burns credits faster than any single complex task.

Everyday setup
  • Model: GPT-5.4-mini (or current stable lower-cost coding model)
  • Speed: Standard
  • Thinking: Low
Everyday anti-pattern
  • Model: GPT-5.5 / latest
  • Speed: Fast
  • Thinking: High or XHigh
Note: Don't make the newest frontier model + Fast mode your everyday default. Fast mode is 1.5x speed but higher credit consumption — that combination is fine for hard problems, but burns more credits than needed for routine coding.

Model Guidance

GPT-5.5 is the frontier model with stronger reasoning and agentic workflows, but GPT-5.4 is sufficient for most tasks.

Model selection by situation
Situation Recommended model Why
Normal coding Lower-cost coding model Strong performance without paying for the latest every time
Small edits Lower-cost coding model Good enough for quick changes
Debugging across files Newer model (within last 2 releases) Needs stronger reasoning
Architecture, migrations, hard bugs Latest model Worth the extra cost for complex tasks
Repeated failures Upgrade model One strong attempt can be cheaper than many weak ones

Speed Guidance

Fast mode is not better coding. It is faster response at higher cost.

Standard vs Fast
Speed Use when Avoid when
Standard Default for almost all coding Don't avoid it just because Fast exists
Fast Live pairing, demos, urgent fixes, short tasks where waiting is the problem Normal development, long repo analysis, large refactors, exploratory debugging

Thinking Level Guidance

Thinking levels by task
Level Use for Token risk
Low Rename, formatting, tiny local changes May miss context
Medium Default coding, tests, normal bug fixes, small features Best balance
High Hard bugs, multi-file reasoning, security-sensitive changes More tokens
XHigh Big migrations, architecture, autonomous repo work Expensive

Gotchas

Things to watch Keep in mind
  • The newest model is not always the best default for cost control.
  • Fast mode increases speed, not coding quality.
  • High thinking can save time on hard tasks, but wastes tokens on simple ones.
  • Many cheap retries can cost more than one well-scoped, higher-effort request.
  • Open files and selected code matter — Codex uses VS Code context, so keep relevant files open to write shorter prompts and get better results.

Best Prompt Patterns

Default setup Everyday codingCopied 
Use the open files only. Make the smallest
safe change. Do not refactor unrelated code.
Explain what changed and list any tests
I should run.
Harder work Complex tasksCopied 
Analyze the relevant files first. Propose
a short plan before editing.
Make the smallest safe change.
Preserve existing behavior unless required.

Bottom Line

Settings summary by task
Task Model Speed Thinking
Everyday coding Stable coding model Standard Medium / Low
Harder coding Stable or latest model Standard High
Urgent coding Any Fast (briefly, then switch back) —
Division of Information Technology

Division of Information Technology

University of Maryland

301-405-1500 itsupport@umd.edu
Instagram Facebook X