Model
Sol Pro reasoning costs more per step than Sol; Luna-class steps cost least.
Reviewed guide · September 1, 2026
Codex limits: coding-agent usage is metered by work, not messages
Codex is included with Plus, Pro, Business, Enterprise and Edu, and it is the feature most likely to exhaust a plan. It is metered by tokens under OpenAI’s rate card, so a vague repository-wide task can cost what fifty small fixes cost. This page explains the metering and how to work with it.
Reviewed: . Figures marked official come from OpenAI’s help center; everything else is labelled as observed or approximate.
How it is metered
Four variables decide what a task costs.
OpenAI: usage “depends on the model, task size, reasoning effort, and amount of work performed”.
Task size
Files read, commands run, tests executed and diffs written all consume tokens.
Reasoning effort
Higher effort settings think longer per step.
Amount of work
An agent that explores a codebase to find the problem spends far more than one told where to look.
By plan
Codex headroom per tier.
| Plan | Codex | Past the allowance |
|---|---|---|
| Free / Go | Not included | — |
| Plus $20 | Included, token-metered | Wait for reset |
| Pro $100 / $200 | Included, 5x / 20x Plus usage | Wait for reset |
| Business | Included per seat | Workspace credits if enabled |
| Enterprise / Edu | Included per contract | Credits and admin controls |
Prompts that save tokens
Scope is the whole game.
| Task | Efficient Codex prompt |
|---|---|
| Bug fix | Expected vs actual behaviour, the failing test command, the file or function you suspect |
| Feature | Acceptance criteria, files to touch, files not to touch, how to verify |
| Refactor | One module at a time, with the test command to run after each step |
| Review | Point at the diff or PR, name the concerns (security, performance, style) |
Make it last
Habits for long coding days.
Plan in chat first
Text is unlimited. Work out the approach in a normal chat, then hand Codex a precise task.
One issue per run
Small scoped runs are cheaper and easier to review than one sprawling instruction.
Name the files
Every file the agent has to discover costs tokens.
Provide the test command
The agent verifies in one cycle instead of guessing.
Lower the effort for routine work
Save the highest reasoning setting for architecture, not renames.
Alternatives
What to use when Codex is capped.
Other coding agents draw on their own plans; the summary you carry across is what matters. None of these are affiliated with OpenAI.
Alternative
Claude
claude.aiStrong writing, analysis and long documents. Has its own five-hour and weekly limits.
AlternativeGemini
gemini.google.comBest inside Google Workspace and for Google-context tasks.
ResearchPerplexity
perplexity.aiResearch answers with web sources and citations.
AlternativeCopilot
copilot.microsoft.comUseful inside Microsoft 365 and Windows.
Multi-modelMultipleChat
multiplechat.aiSeveral models in one workspace so one provider’s cap does not end your day.
Disclosure: MultipleChat is built by the team that runs this site. It is listed as one option among several; nothing on this page is paid placement.
Official sources
Where to verify the current rule.
Checked September 1, 2026. OpenAI changes plan rules without notice; these pages and your own in-app counter are authoritative.
OpenAI Help: Codex rate card
Token-based Codex usage, credits and plan coverage.
OpenAI Help: GPT-5.6 in ChatGPT
Which reasoning options each plan gets, unlimited text on Free and Go, and what happens at a reasoning limit.
ChatGPT pricing
Current plan names, prices and what each tier includes.
OpenAI Help: Flexible pricing for Business, Enterprise and Edu
Workspace credits and how paid usage continues after an allowance.
FAQ
Codex limit questions answered.
Is the Codex limit the same as the ChatGPT limit?
No. Text chat is unlimited; Codex is metered by tokens under OpenAI’s Codex rate card, and usage depends on the model, task size, reasoning effort and the amount of work performed.
Which plans include Codex?
Plus, Pro, Business, Enterprise and Edu. Pro has more headroom than Plus; Business and Enterprise can extend usage with workspace credits.
How many Codex tasks do I get?
OpenAI does not publish a task count because tasks vary enormously. A one-file fix and a repository-wide refactor with tests can differ by orders of magnitude in tokens.
What uses the most Codex allowance?
Long autonomous runs that read many files, high reasoning effort, repeated test cycles, and vague prompts that make the agent explore.
Can Support reset Codex usage?
No. OpenAI states that Support does not reset ChatGPT or Codex usage limits.
Codex CLI or the ChatGPT Codex tab — same limit?
They draw on the same plan allowance and rate card. Check the in-product usage messages; they are the source of truth.