oid and user_address, start March 2026. All of it over REST, WebSocket, and Parquet export.
Use core perp symbols such as BTC here. Use Spot for pairs like HYPE-USDC, HIP-3 for builder markets, and HIP-4 for outcome markets. Lighter is a separate venue with its own order book routes.
Order book routes
L2, L3, and L4 route families.
Order book depth
What L2, L3, and L4 mean.
What Order Book Data Is Available
| Depth | Earliest | Detail |
|---|---|---|
| Native L2 (20 levels/side) | April 2023 | Aggregated price levels (px, sz, n); more than 24 billion snapshots |
| Full-depth L2 (from L4) | March 2026 | Any depth via depth=; aggregated from the order-level book |
| L4 order-level | March 2026 | Every resting order with its oid and user_address, plus diffs and reconstruction |
| L3 on Lighter | March 2026 | Lighter’s order-level book; a separate venue, see Lighter |
L2, L3, and L4 Order Book Data
L2 is aggregated depth: each price level with total size and order count. L4 is order-level: every resting order with its ownoid and user_address. Use L2 for spread, depth, and slippage history. Use L4 when the job needs queue position, per-order detail, or book reconstruction. Hyperliquid core, Spot, HIP-3, and HIP-4 all expose L4; Lighter exposes L3.
Route Map
| Need | Route family to inspect first |
|---|---|
| Current or timestamped depth | /v1/hyperliquid/orderbook/{symbol} |
| Historical depth snapshots | /v1/hyperliquid/orderbook/{symbol}/history |
| Order-level (L4) history | /v1/hyperliquid/orderbook/{symbol}/l4/history |
| Order lifecycle and flow | /v1/hyperliquid/orders/{symbol}/history, /v1/hyperliquid/orders/{symbol}/flow |
Which Depth Do I Need
Pick depth by the downstream job, then pull one bounded window before widening.Backtesting and slippage
L2 depth snapshots give spread, depth, and slippage history for one symbol and window.
Execution and microstructure
Order-level and lifecycle routes give placement, cancel, and fill detail for queue and flow analysis.
Gate downstream use
Check Data quality, freshness, and request IDs before the output feeds research, models, or alerts.