How many tokens does a Codex agent run use?

A typical Codex autonomous task burns 60,000–120,000 tokens (input + output combined). Long agent runs with file reads and multiple tool calls can exceed 250,000 tokens. A simple chat message is closer to 5,000 tokens.

How long is the Claude Code 5-hour window?

Claude Code uses a rolling 5-hour window starting from your first message. After 5 hours, your token budget resets. Heavy agent users on Pro often exhaust the budget in 30–60 minutes, then must wait for the window to roll over.

Can I see my remaining token budget?

Not directly from Anthropic's dashboard in real time. SessionWatcher ($2.99 one-time) shows your live token consumption, percentage of window used, and minutes until reset right in your macOS menu bar.

Token-to-Time Estimator

Codex agent eats ~85k tokens per run. How many fit in your 5-hour window? Dial in your usage and find out.

1Your plan

2Tokens per agent run85k

Quick editLong agent

3Runs per hour2

Slow & carefulContinuous

In a 5-hour window on Claude Max 5×

Runs before lockout

0h 32m

Until lockout

944%

Of window used

At 2 runs/hour you'll exhaust Claude Max 5×'s budget in 32 minutes — well before the 5-hour window resets. Upgrade or batch your runs.

Find a plan that fits Score my lockout risk

Typical token costs

Single chat message~5k tokens
Claude Code edit + run~30k tokens
Codex autonomous task~85k tokens
Long /loop or babysit run200k+ tokens

Why token math beats message math

Counting messages is a lie. A one-line chat message and an autonomous agent run both count as “1 message” in your head — but the agent burned 20× the budget. The 5-hour window cares about tokens, not turns.

What the calculator assumes

We use published 5-hour token caps per plan, divide by your tokens-per-run input, and compute the resulting runtime at your chosen run-rate. The 25% safety margin you saw on the plan calculator is not applied here — this shows the raw ceiling so you can see exactly when you'd hit it.