---
title: PubChem rate limits
description: 'Pace PubChem at 3-5 requests per second, 1-2 for structure searches, and back off from 1-2 seconds to 30 on a 429 while honouring Retry-After.'
---

## PubChem rate limits

Stay within PubChem’s throttling while keeping results responsive. Inside chat, Ivy paces PubChem requests for you; the guidance below is for scripts that call PubChem directly, on your own machine or in the chat's [code workspace](/docs/sandbox).

## Recommended pacing

- Page requests: 3–5 requests per second per client is typically safe.
- Structure searches: treat as heavier; pace at 1–2 requests per second.
- Large exports: split into pages and download sequentially.

## Backoff strategy

- Use exponential backoff starting at 1–2 seconds, doubling to a max of ~30
  seconds.
- On HTTP 429, respect `Retry-After` if provided; otherwise wait at least 5
  seconds.
- Avoid concurrent exports for the same query; reuse cached results instead.

## Caching behavior

- The platform may cache recent queries; cached responses return immediately with a
  timestamp.
- Re-run the same query only when you need fresher data; otherwise reuse the
  cached export.
- Log whether a response came from cache to interpret latency.

## Monitoring tips

- Track error rates and 429s per client to tune pacing.
- Record query type (text/identifier/structure), filters, and page size to
  correlate with throttling.
- Keep request IDs in logs so support can trace failures.

## If you’re repeatedly throttled

- Lower concurrency and increase delay between requests.
- Narrow queries (add filters) to reduce result size.
- Prefer identifier or InChIKey searches over broad synonym queries for heavy
  workloads.