Computing Units

Every Open API endpoint advertises a per-call cost called a Computing Unit (CU). CU is how Pendle meters the work each request does and enforces fair-use against high-frequency consumers.

Budget

Two fixed-window buckets per IP, both must have capacity for a call to succeed:

Window	Limit
1 minute	200 CU
1 week	400,000 CU

Every successful call deducts its x-computing-unit cost from both buckets. The minute bucket refills every 60s; the weekly bucket refills every 168h.

When either budget is exhausted the API returns HTTP 429 with body { "statusCode": 429, "message": "ThrottlerException: Too Many Requests" }. Wait for the next window before retrying.

Endpoint-Specific Throttles

A few endpoints have additional per-IP throttles on top of the CU budget:

Endpoint	Extra throttle
GET /v1/indicators/export	3 requests/minute/IP

Higher Rate Limits via API Key

If your integration needs more than the per-IP free-tier limits, paid plans are available through the Boros API Dashboard.

Getting an API key

1. Go to https://api-boros.pendle.finance/dashboard
2. Connect your wallet and sign the verification message to log in
3. Choose a plan and create a new API key

Save the generated key securely — you'll need it to authenticate your requests.

Using your API key

Include your key as a Bearer token in the Authorization header on every request:

Authorization: Bearer <API_KEY>

Example:

curl -i -H "Authorization: Bearer your_api_key_here" \
  https://api-boros.pendle.finance/apis/v1/markets

Authenticated requests are billed against your plan's higher limits instead of the per-IP free-tier budget.

Where CU is Reported

Surface	Format
OpenAPI spec	x-computing-unit extension on every operation. Static endpoints publish a number ("1", "2", "20"); dynamic endpoints publish a number with a + suffix ("1+", "3+").
HTTP response	x-computing-unit response header on every successful call. For dynamic endpoints, the header reports the actual CU charged for that specific request.

The redesigned interactive docs at api-boros.pendle.finance/apis/docs display the published cost next to every endpoint.

Static vs Dynamic Cost

Static ("N")

The endpoint always charges N CU per call regardless of parameters. Examples:

Endpoint	CU
GET /v1/assets	1
GET /v1/markets	2
GET /v1/markets/order-book	2
GET /v1/on-chain-events	3
GET /v1/indicators/export	20
Any single calldata-builder POST	1

Dynamic ("N+")

The endpoint charges at least N (the minimum, billed for an empty / default request) and scales with the work performed. The actual cost is reported in the response header. Common scaling axes:

• Page size — list endpoints charge proportionally to limit. Example: GET /v1/events/liquidation-events is max(1, ceil(limit / 200)).
• Batch size — multi-item endpoints charge 1 CU per 10 items. Example: POST /v1/funding-rate/settlement-summary is max(1, ceil(settlementSummary.length / 10)); POST /v1/calldata-builder/agent/place-orders is max(1, ceil(orderRequests.length / 10)).
• Number of indicators — GET /v1/indicators charges 1 CU per indicator listed in select. A udma:7;30 entry counts as one indicator regardless of how many windows are packed in.
• maker lookup flag — GET /v1/incentives/maker-incentives/campaigns/{marketId} is 1 CU without maker, 2 CU with maker.

The minimum-and-grow pattern means callers paying only for what they use — passing fewer items / indicators is cheaper.

Reading the Header in Practice

const res = await axios.get(`${BASE_URL}/v1/indicators`, {
  params: { marketId: 1, timeFrame: '1h', select: 'u,fp,fgi,udma:7;30' },
});
const cuCharged = Number(res.headers['x-computing-unit']);  // e.g. "4"

For dynamic endpoints, you can sum the header over a window to reason about total compute spent.

Tips

• Static endpoints are cheap. The vast majority of read endpoints are 1 CU. Build dashboards against them without worrying about cost.
• Batch where possible. POST /v1/accounts/market-acc-infos accepts up to 100 marketAccs in one request; GET /v1/markets/by-ids and GET /v1/amm/states likewise accept comma-separated IDs. One batched call is far cheaper than N individual calls.
• Prefer streaming for long histories. For multi-day indicator windows, GET /v1/indicators/export is a flat 20 CU regardless of the indicators selected — much cheaper than hammering /v1/indicators repeatedly.
• Calldata batching is cheap. A place-orders request with up to 10 orderRequests[] entries is 1 CU; cost grows by 1 per additional 10 entries.