QuotaClear

A PASTED card alone is not a measured Claude usage limit. In our Free-account check, a 5,000-character paste became a card before we sent the message. Inspect the input and the actual limit message separately; the card is not evidence of a quota deduction or a token saving.

Is a PASTED Card a Claude Usage Limit Warning?

Tested November 7, 2026 · Claude web composer · Account QC-FREE-01, an internal test label

Here is the narrow question we tested: what happens to a long clipboard paste before the message is sent? This is a composer check. It is not an allowance benchmark, a Pro test, or a comparison of model answers.

What appeared in our Free account

The account showed Free. The model selector showed Sonnet 5.5, with medium effort selected. We did not press Send during this check.

Each input was the ASCII sequence 0123456789 repeated to the character count below. Read the rows as one controlled series, rather than separate claims about a universal cutoff.

Observed in the new-chat composer on November 7, 2026
Characters pastedVisible input stateCard label
1,000Text stayed in the input boxNo card
5,000PASTED card; input box empty1 line
10,000PASTED card; input box empty1 line
20,000PASTED card; input box empty1 line

In this series the presentation changed between the tested inputs. We did not locate the exact transition point. Do not turn these observations into a fixed character allowance for every account, browser, or future version. In particular, do not use the card’s line label as a token count.

Our composer check: one thousand characters stayed inline; five thousand and larger tested pastes became cards; editing restored the complete inputs.
Our observed input states. This diagram is not a quota meter.
PASTED is an input state, not a measured quota deduction.
One thousand characters stayed inline; five thousand, ten thousand and twenty thousand became cards.

Was the full text still there?

For each of the three cards, we clicked its edit control. The text returned to the composer: respectively 5,000, 10,000, and 20,000 characters. A direct comparison of the restored text with its original input matched exactly in every one of those three checks.

That is an edit-and-recovery observation, not a claim that a model read the content correctly. We did not measure input tokens, context occupancy, remaining messages, or a before-and-after usage percentage. Those fields remain unmeasured.

Editing each of the three cards restored text exactly equal to its original input.

One incidental draft-retention observation

After the initial 1,000-character control, we clicked New while already on the new-chat page. That unsent draft remained in the composer. This was an incidental observation from one navigation, not a preregistered trial or a claim about what New does from an existing conversation. Inspect the actual input before assuming it has been cleared.

Inspect your input and treat any actual limit message separately from the card label.

Check your paste before diagnosing a limit

  1. Look for an inline prompt or a PASTED card after pasting.
  2. Open the card’s edit control if you need to inspect the text. Check its beginning and end against your original.
  3. Review the complete input before sending it. Keep any actual limit message separate from the attachment label.

Use this table to recognize the input state we observed. It cannot tell you how much allowance a later request will consume, whether that request will be accepted, or whether converting the input changes its cost.