AI Agent - Aug 2, 2026

ChatGPT Work vs Chat vs Codex: Which Workflow to Use

Quick answer

  • Use standard Chat for conversational questions, exploration, and lighter drafting.
  • Use ChatGPT Work for longer, multi-step research, analysis, and finished business artifacts across authorized chats, files, workspace resources, and connected systems.
  • Use Codex when the primary work object is software and the job needs repository context, file changes, commands, tests, diffs, or pull-request workflows.

The surfaces can overlap. Choose by the system of record, execution environment, deliverable, and governance burden—not by which one can produce the most fluent answer.

Decision matrix

DimensionStandard ChatChatGPT WorkCodex
Best starting pointQuestion or conversationDelegated multi-step outcomeCodebase or development task
Typical contextMessages, selected files, enabled toolsChats, files, workspace resources, connected systemsRepository, local or cloud environment, commands, developer tools
Typical outputAnswer or draftDocument, presentation, analysis, spreadsheet, Site, reviewed artifactPatch, diff, test result, code explanation, PR work
Runtime concernTool-specificSurface, plugin, source, action, sharing, scheduleFilesystem, network, sandbox, approvals, worktree, shell
Review focusFactual and editorial reviewSource lineage, permissions, business acceptance, destinationDiff, tests, security, behavior, repository conventions

Choose standard Chat

Use Chat when a person remains in the loop turn by turn and the output is primarily conversational. It is a good starting point for clarifying a question, drafting a short passage, or deciding whether a larger delegated task is well specified.

Do not assume that standard Chat and Work share the same usage accounting. OpenAI’s current enterprise documentation says standard Chat usage is separate from the credits and limits shared by Work and Codex.

Choose ChatGPT Work

Use Work when the task crosses several authorized sources or needs a finished, review-ready deliverable. Define the source boundary, decision, format, acceptance tests, data policy, action permission, and sharing state before delegation.

Connected systems preserve native permissions, but aggregation still changes risk. Review citations, calculations, missing sources, freshness, recipient scope, and any external action.

Choose Codex

Use Codex when software is the central object. Repository work benefits from project instructions, controlled filesystem and network access, commands, tests, diffs, and development-specific review. Local and cloud Codex runs have different environments and permissions; choose intentionally.

Codex can also research, write, analyze, and create artifacts, but that overlap does not erase its codebase-first execution and review strengths.

The ChatGPT Work guide covers source and deliverable controls. The scheduled-tasks guide covers repeatable execution.

Frequently asked questions

When should I use ChatGPT Work instead of standard Chat?

Use Work when the job is longer and multi-step, needs authorized files or connected systems, or should return a review-ready document, presentation, analysis, spreadsheet, or Site. Use standard Chat for conversational questions and lighter drafting that does not need delegated execution.

When should I use Codex instead of ChatGPT Work?

Use Codex when the primary system is a codebase or development environment and the job needs repository inspection, edits, commands, tests, diffs, or pull-request work. Use Work when the primary outcome is research, analysis, or a finished business artifact across authorized sources.

Do ChatGPT Work and Codex use the same credits?

OpenAI’s current admin FAQ says ChatGPT Work and Codex share pricing, credits, and usage limits, while standard Chat usage is separate. Exact entitlement and consumption still vary by plan, workspace, model, task duration, context, tools, and output.

Bottom line

Pick the smallest surface that matches the job: Chat for conversation, Work for delegated finished work, and Codex for software execution. Recheck permissions, usage, and governance whenever a task crosses surfaces.

Official sources

Source check: August 2, 2026. Verify current plans, usage, surfaces, runtime controls, and rollout state.