TokenMaxer

Codex limits / live utility

Codex usage limits, context window, and token visibility

Use this when Codex says the limit is close, context starts compacting, a configured window does not stick, or a team needs to decide whether Codex or Claude Code will last through a real coding session.

720/mocodex usage limits $67.53context-window CPC 2026-06-20last checked
Direct answer Codex does not have one universal public limit.

Plan allowance, task size, execution surface, context pressure, and model/client behavior all affect how quickly you hit the wall.

Resolve your case

Limit resolver

Pick the plan and the symptom.

Free / Go + Limit banner Limited trial access

Open the Codex usage page first. If the current turn is already running, Codex can usually continue that turn subject to fair-use limits, then you use the banner/page options after the turn.

Use Codex to test fit. If you are hitting limits while doing real work, the answer is usually plan upgrade or smaller tasks.

Plan limits are not message counts.

6 rows
Plan Official read Best fit Practical limit read Source
Free Limited trial access Testing Codex before a paid plan. Expect the narrowest allowance. Check the limit banner when it appears. S2
Go Limited trial access Low-frequency trials where Codex is not the main coding tool. More access than Free, still not a sustained daily coding plan. S2
Plus A few focused coding sessions each week Short bugfixes, small features, and focused repo work. Usage depends on task size and where the task runs. Credits may be available when near limit. S1 / S2
Pro 5x or 20x more usage than Plus Longer daily sessions and heavier coding workflows. Still not literally unlimited. Long context, large repos, and cloud tasks burn allowance faster. S1 / S2
Business Codex Pay as you go based on usage Teams that want Codex without fixed seat assumptions. Treat this as usage-metered. Pair with budgeting, analytics, and admin policy. S2
Business / Enterprise Workspace controls and Codex usage visibility Managed teams that need auditability and policy controls. Use workspace analytics and Compliance API where available; local and cloud usage are separate surfaces. S1 / S2

Context window: official number vs working window.

API != app
Official

GPT-5.3-Codex API model

400,000 context window; 128,000 max output tokens

Good official anchor for API/model capability. Do not assume every Codex client exposes the full practical window.

S3
Mixed

Codex app / CLI effective session

Varies by model, client, config, and surface

The effective working window can be lower than the model headline when the client compacts, caps, or ignores unsupported config.

S1/S4
Observed

Large custom context settings

Config knobs exist, but unsupported values may not behave as expected

A GitHub issue reports a roughly 1M setting being reduced or ignored around 258k in one Windows/config case.

S4
Official

Context pressure

Long sessions and larger codebases consume more allowance

A message is not a fixed unit. File reads, pasted content, instructions, and conversation history all change the burn.

S1

Context pressure estimator

Estimate how close the session is to the wall.

Estimated context pressure 54%

Healthy. Keep working, but compact before a large direction change.

Official API model context window.

Where to look when usage disappears.

5 surfaces
Codex usage page
Shows
Plan-level usage status and options when near or at a limit.
Misses
Per-task token breakdown.
Action
Use first when a banner or limit warning appears.
Limit banner
Shows
Immediate state when the current plan or window is constrained.
Misses
Why a specific turn consumed so much.
Action
Follow the banner for credit, reset, upgrade, or wait options.
Codex CLI status
Shows
Local-session state where supported.
Misses
Unified history across every Codex surface.
Action
Run status checks before blaming the plan.
Local session files
Shows
Inspectable local run artifacts when present.
Misses
Official billing truth.
Action
Use for debugging and estimating; do not treat as billing record.
Compliance / Enterprise Analytics API
Shows
Workspace-level Codex usage for eligible enterprise setups.
Misses
Personal-plan users.
Action
Use for team governance and budget audit.

When to compact, restart, or change plan.

Decision flow
Same goal Compact.

The task is still stable, but the session is carrying too much history. Compact before the next large read.

New goal Start fresh.

If Codex explored the wrong files or the task changed, a fresh prompt with narrow paths usually beats preserving clutter.

Config mismatch Verify the cap.

If a large context setting does not stick, separate model limit, client limit, and unsupported config before debugging.

Plan wall Check usage.

If the banner says you are constrained, use the plan options first. Context tricks do not create allowance.

Codex limits FAQ

4 answers
Does Codex have fixed message limits?

Not in a simple public-message-count sense. OpenAI says Codex usage counts toward agentic usage and varies with task size, complexity, and where work executes.

Is the Codex model context window the same as the app context window?

No. The API model page is an official model capability. The Codex app or CLI may have effective client behavior, compaction, or config limits that make the working window feel different.

What should I do when Codex starts compacting context?

Compact if the goal is stable and the session still has the right files. Start fresh if the task direction changed, the session read too much unrelated code, or the important state is easier to restate than preserve.

Can I see exact token usage per Codex conversation?

Some usage surfaces exist, especially for plans and enterprise analytics, but per-conversation visibility is fragmented. This is why the local session and dashboard checks are separate in the matrix.

Source ledger

5 checked
S2

Codex Pricing

ChatGPT - Plan positioning for Free, Go, Plus, Pro, Business, and Enterprise Codex access.

S3

GPT-5.3-Codex model page

OpenAI API docs - Official model context window, max output tokens, and token pricing for GPT-5.3-Codex.