Docs
Limits and units
Each plan has a monthly allowance of searches, plus short-term limits that keep the service fast for everyone.
What counts as a search
- search_papers (first page)
- 1 search
- search_papers with cursor (next page)
- free
- expand_citations
- 1 search
- any call that returns an error
- free
A search is reserved when a call starts and charged only if it succeeds. Failed or cancelled calls give it back.
Limits per plan
| Plan | Searches / month | Burst | Sustained rate | Parallel calls |
|---|---|---|---|---|
| Free | 25 | 5 calls | 1 call every 2 s | 2 |
| Pro | 1,000 | 10 calls | 1 call every 1 s | 3 |
| Max | 5,000 | 10 calls | 1 call every 1 s | 4 |
When the allowance resets
Free plans reset on the first day of each calendar month (UTC). Paid plans reset at the start of each billing month; annual plans still reset monthly. Unused searches do not roll over.
What happens at a limit
- Monthly allowance used up: calls return
BUDGET_EXCEEDEDwith your usage, the reset date and an upgrade link. Nothing is charged as overage. - Too many calls in a row, or too many at once: calls return
RATE_LIMITEDwithretry_after_s. Claude can wait and retry.
Response size
Each hit carries its full abstract, so a page of 10 hits is typically 4,000 to 8,000 tokens by our conservative count and at most about 9,000, under Claude Code’s 10,000-token warning. Hits that do not fit move to the next page. Up to 25 hits per page; when truncated is true, the next page is one call away and free.