Hyperliquid rate limits
Hyperliquid's info API needs no key, which is why people assume it is unlimited. It is not: every request spends from a shared 1,200 weight-per-minute budget per IP address, and the endpoints you need most for history work are the most expensive ones. This page documents the weight table, the math for fills exports, the 429 behavior we measured crawling a 45k-fill wallet, and the official escape valve.
- Aggregate budget: 1,200 weight per minute per IP across all REST calls.
- History endpoints (
userFills,userFillsByTime,userFunding,fundingHistory,recentTrades,historicalOrders, TWAP/delegator endpoints…) cost base 20 + 1 weight per 20 items returned — a full 2,000-fill page costs 120 weight. - That caps deep pagination at ~10 full calls/min ≈ 12,000 fills/min theoretical ceiling — our 45,147-fill month needs ≥4 minutes of crawling.
- Overage returns HTTP 429; measured behavior is transient failure, not a hard wall — retry with back-off. Sustained violations can get an IP banned.
- Bulk alternative: requester-pays S3 buckets (node_fills_by_block, node_trades, misc_events_by_block) — see the bulk download guide.
The weight table (October 2026 docs)
| Request | Weight | Surcharge |
|---|---|---|
info: l2Book, allMids, clearinghouseState, orderStatus, spotClearinghouseState, exchangeStatus | 2 | — |
info: userRole | 60 | — |
info: all other documented calls | 20 | — |
info: userFills, userFillsByTime, userFunding, fundingHistory, recentTrades, historicalOrders, nonUserFundingUpdates, twapHistory, userTwapSliceFills(ByTime), delegatorHistory, delegatorRewards, validatorStats | 20 | +1 per 20 items returned |
info: candleSnapshot | 20 | +1 per 60 items returned |
exchange actions | 1 + floor(batch/40) | — |
explorer calls | 40 | blockList: +1 per block |
| EVM JSON-RPC (rpc.hyperliquid.xyz/evm) | — | separate 100 req/min cap |
Note the surcharge mechanism: it is the response size that is billed, not the request. An empty page costs base weight only — which is why the correct crawl pattern is "page until empty", not "page until error".
The math for a fills export
- 1 call of
userFillsByTimereturning a full 2,000-fill page: 20 + (2000/20) = 120 weight - Budget 1,200/min ÷ 120 = ~10 full calls per minute ≈ 12,000 fills/min
- A 45,147-fill month: ≥ ~4 min of sustained crawling. Our scanner's measured crawl of that wallet took minutes, matching the cap.
- Funding is lighter:
userFundingpages at 500 items → 20 + 25 = 45 weight per full page.
Everything else sharing your IP counts against the same 1,200 — a second tab running a scan, a bot on the same egress, your own retry loop. Budget for it.
What a 429 actually means (measured)
Deep-crawling a wallet with 80k+ fills we hit stretches of failed responses mid-history. The pattern that mattered: the failures were transient — retrying the identical window after ~1.5s returned data again, and fills continued for months past the failure point. Two rules fall out of this:
- Empty array = end of data. An error ≠ end. Only stop on a clean empty response.
- Back off and re-ask the same window before concluding anything — a 429 is billing, not an answer.
Per-address limits: a separate budget
Besides the per-IP budget, exchange actions (orders/cancels — not reads) are capped per address, scaled by trading volume. You can query the budget itself:
POST https://api.hyperliquid.xyz/info
{"type":"userRateLimit","user":"0x…"} Measured response on a $18.9B-volume wallet (October 20, 2026):
{"cumVlm":"18928394140.0200004578",
"nRequestsUsed":6440,
"nRequestsCap":18928404140,
"nRequestsSurplus":0} The cap tracks cumulative volume with a ~10k-request floor — irrelevant for read-only tax exports, but the field to know if you ever write trading code.
Websocket and EVM caps
- WS: 10 connections per IP, 30 new connections/min, 1,000 subscriptions, max 10 unique users across user subscriptions, 2,000 outbound messages/min, 100 in-flight posts.
- EVM JSON-RPC on
rpc.hyperliquid.xyz: 100 req/min — third-party HyperEVM RPC providers have their own (usually richer) quotas.
The official escape valve: S3
The docs' own answer for large batch needs is "use the S3 bucket instead" — requester-pays buckets holding fills by block, trades, non-trade events (funding, staking, transfers) and L2 book snapshots. Hundreds of GB of raw history, billed to your AWS account at cents per GB. Routes and bucket layout: Hyperliquid data download guide.
Your own history, paced for you
The scanner paginates within the weight budget automatically — fills, funding and ledger, grouped per tax year. Free, in your browser.
Scan my walletCommon questions
What is the Hyperliquid API rate limit?
REST requests share an aggregate budget of 1,200 weight per minute per IP address. Most info calls cost 20 weight, but the history endpoints (userFillsByTime, userFunding and friends) add a surcharge of 1 weight per 20 items returned — so a full 2,000-fill page costs 120 weight and you can only afford about ten of those per minute.
What happens when I exceed the limit?
Requests start returning HTTP 429 instead of data. In our deep crawls the failures are transient — the same window succeeds after a short back-off — but sustained overage can get an IP banned. Retry the same window; never treat an error as the end of your data.
How do I get around the limit for bulk exports?
For your own tax history you cannot go around it, only pace to it — roughly 10 full fills calls per minute. For bulk market data the official route is the requester-pays S3 buckets (node_fills_by_block, node_trades, misc_events_by_block, market_data snapshots), not the API at all.