Low-Latency Trading Solutions
Cart
Historical order book data

Historical Order Book Data: Level-2 Depth and Market Depth History for Crypto

The order book is the part of a market that no candle feed keeps. Every day file in the archive carries the full Level-2 evolution of one instrument — one snapshot, then every incremental update with the venue timestamp and our co-located receive timestamp in integer nanoseconds — for 340,065 instruments across 35 venues, February 2021 – present. Replay it and the book is there at any instant: depth level by level, spread, queue position, imbalance. Any instrument-day is €1, bought as a guest with instant download.

Venues
35
Instruments
340,065
Coverage
February 2021 – present
Recorded
7.85 PB
Depth-only venues
6
Per day
€1
Order book

Historical Order Book Data — L2 Updates, €1

Level-2 history is recorded as the venue sends it: a type-0 snapshot resets the book, then type-1 updates change one price level at a time (quantity 0 removes the level). The free Python reader replays a file into a fixed-grid order-book dataset in one command (book FILE --every 1s --depth 20 --out book_1s.parquet), and the browser viewer replays any sample or purchased day tick by tick.

Top-of-book is a separate best-bid/ask stream where the venue publishes one (type 6) and is derived by replay where it does not: BitMEX, Coinbase, Kraken spot, Kalshi, Polymarket, predict.fun, OG.com record depth only. That is why the flat bbo.csv.gz export exists only for venues with a BBO stream — on the depth-only venues the reader rebuilds the top of book locally from the L2 updates, which is exact, not sampled.

What the depth history answers that trades alone cannot: how much size sat at each level before a move, how long orders queued before they filled or cancelled, how the spread behaved through a liquidation cascade, and whether a market impact model is calibrated on the book that actually existed. The order-book dataset guide turns the replay into features and labels.

Back to the historical-data overview →

Scale

How big is a day of order book history?

ExampleCompressedEvents
Binance BTCUSDT perpetual, one day1–2 GBtens of millions
BitMEX XRPUSDT perpetual, one day59 MB1.87 M (1.83 M book updates)
Kalshi event contract, one day≈ 94 KB on averagethousands
Polymarket event contract, one day≈ 353 KB on averagetens of thousands

Examples from the format reference; the order-book updates are the bulk of every file.

Depth-only venues

Venues without a separate best-bid/ask stream

These venues publish Level-2 depth but no top-of-book message — the free reader derives the best bid and ask by replaying the book, exactly, not sampled. No flat BBO export exists for them.

Live snapshot

Spread and depth per venue — last 24 hours

A live, display-only snapshot of the BTC/ETH probe basket: how tight and how deep each venue's book runs right now, from the same minute statistics the archive derives. It changes hourly and is not part of the structured data.

VenueMedian spreadTop-1 depthTop-20 depthProbed instruments
Kraken Futures1.39 bps$9.6K$1.5M5
BingX0.23 bps$51.2K$1.4M10
Binance Coin-M0.17 bps$23.9K$1.1M5
HashKey MENA0.65 bps$34.2K$619.5K4
Gate.io Future1.05 bps$9.7K$526.7K5
Okx0.32 bps$22.8K$524.2K10
Binance USDT-M0.17 bps$93.9K$511.8K5
Orderly2.07 bps$7.2K$490.7K5
Bitget0.16 bps$16.2K$464.1K10
HashKey Exchange2.05 bps$30.0K$434.9K4
Mexc1.45 bps$6.8K$433.2K5
Hyperliquid0.22 bps$37.0K$376.7K5

Live 24 h snapshot of the BTC/ETH probe basket per venue — median bid-ask spread and mid-side total notional over the top 1 and top 20 levels ((bid + ask) / 2 × levels), from the minute statistics of the archive. Changes hourly; the full dashboard is the liquidity analyzer.

Recorded activity

Venues with order book history, ranked by recorded turnover

Every venue below carries the Level-2 book next to its trades — 31 venues with activity in the 30 days to 2026-11-17.

#VenueTurnover · 30d (USD)ShareTrades · 30dAvg tradevs prior 30d
1Binance USDT-M$1.8T33.3 %5.2B$335+4.5 %
2Okx$768.3B14.6 %920.2M$835+3.3 %
3Bybit$450.8B8.6 %1.3B$344−3.6 %
4Gate.io Future$435.4B8.3 %438.3M$993+13.2 %
5Deribit$292.2B5.5 %74.6M$3,917+157.3 %
6Bitget$273.1B5.2 %527.3M$518+8.3 %
7Hyperliquid$272.7B5.2 %166.4M$1,639+20.7 %
8Whitebit$227.0B4.3 %431.8M$526−28.6 %
9Binance Spot$218.5B4.1 %804.4M$272−29.8 %
10BingX$144.6B2.7 %2.8B$52−14.4 %

USD turnover and trade counts per venue over the 30 days to 2026-11-17, summed from every trade print in our tick archive (31 venues with activity; share = of $5.3T total). Not venue-reported volume.

All venues, month by month and by instrument class →

What's in the file

Message types behind order book data

TypeMessageWhat it carries
0Order-book snapshotLevel-2 book state; a new snapshot resets the book
1Order-book updateIncremental L2 change per price level (quantity 0 removes the level)
6Top-of-book (BBO)Best bid/ask stream where the venue publishes one (not on BitMEX, Coinbase, Kraken spot, Kalshi, Polymarket)

All message types of a day file →

How to

How to download order book data

  1. Find the instrument in the Data Shop (search by pair or venue) or browse by exchange, by asset or by trading pair.
  2. Try a free day first: the full-day samples on the downloads page are the exact shop files, no account needed — parse one before you buy a month.
  3. Pick the UTC days in the instrument's calendar and check out — card with invoice, or Premium and pack credits (1 credit = €1 = one instrument-day). Many instruments at once: "Select to buy" in the instrument scanner adds a shared date window.
  4. Replay: book FILE --every 1s --depth 20 with the free reader for a fixed-grid Parquet at any depth (20 shown — the file holds the depth of its feed), or open the day in the browser viewer to scrub the ladder and tape visually. Trades and (where published) BBO also export flat as ?format=trades.csv.gz / ?format=bbo.csv.gz.
  5. Download from the order page (single files), your account (batch ZIP of everything you own) or scripted via the MCP server and the Agent Skill. Days in deep archive restore in about 12 hours.
Usage

Research, backtesting and model training

Purchased data is for your own use — backtesting, research, analytics and training your own models, in-house or in your own products; the trained models and derived data you build from it are yours. The raw files and tick streams themselves may not be redistributed, resold or published. The full terms are on the data license page.

Why CryptoStruct

Why buy order book data here

Four things every page on this site is built on — and the reason the numbers above exist at all.

We record everything

The complete public feed of each venue as it was published: every Level-2 snapshot and update, every trade with its aggressor side, every quote, funding, mark-price and liquidation event — for every instrument the venue lists, every UTC day since we added the venue. Where a venue offers more than one order-book feed, we take the one with the lowest latency, at the depth that feed carries. Nothing sampled, no on-demand capture.

Institutional grade

Captured co-located at the venue with the exchange timestamp and our receive timestamp in integer nanoseconds, an event-id chain that makes any gap visible, and one normalized schema across 35+ venues — the same capture our own high-frequency trading engine and enterprise feeds run on.

€1 per instrument-day

Any instrument-day is €1, series-day bundles start at €1 — no subscription, no minimum order, no tiers to unlock. Credit packs lower the effective price and never expire, and free full-day samples from the major venues let you test first.

Self-service for everyone

Pick the days in the Data Shop, pay by card as a guest and download immediately — no sales call, no enterprise contract, no KYC. Coding agents buy the same files through the MCP server, and the free Agent Skill teaches them the format.

Agents: /llms.txt · MCP server /mcp · every page as markdown via Accept: text/markdown

FAQ

Order book data — FAQ

Is the order book history complete or sampled snapshots?

Complete. A day file holds one initial Level-2 snapshot followed by every incremental update the venue published, each with its exchange timestamp and our receive timestamp in nanoseconds. Nothing is sampled on the way to disk, and an event-id chain makes any gap visible — the free reader's audit command reports chain breaks and crossed book states.

How deep is the Level-2 data — how many price levels?

As deep as the order-book feed we record. Where a venue offers more than one order-book feed, we take the one with the lowest latency, at the depth that feed carries. On Bybit that feed carries the top 50 levels per side, on OKX the top 400 levels per side. Nothing is cut on the way to disk — a top-20 grid is the common research cut, and the reader takes --depth as a parameter. Check the current docs of the venue for the feeds it offers.

Which venues have no separate best-bid/ask stream?

BitMEX, Coinbase, Kraken spot, Kalshi, Polymarket, predict.fun, OG.com record depth only — there is no type-6 BBO message in their files and no flat BBO export. Top-of-book on these venues comes from replaying the order-book updates, which the free reader does locally and exactly. Every other venue carries the BBO stream in the same file next to the depth updates.

Can I get Binance order book history for a specific day?

Yes — every instrument-day is sold individually, so one day of BTCUSDT on Binance USDⓈ-M or Binance spot is one file for €1, with the order-book stream, trades, BBO, funding and liquidations inside. Pick the day in the shop calendar; older days in deep archive restore in about 12 hours.

How do I turn the order book history into a dataset?

Replay the file on a time grid: the reader's book command emits the top-N levels every N milliseconds or seconds as Parquet, and from that grid come the usual features — depth imbalance, microprice, spread, queue sizes, book-pressure labels. The dataset guide walks through the pipeline with leakage-safe labels; the training-data page covers sizes.

Is this Level-3 (market-by-order) data?

No — the archive is Level-2: one row per price level with the venue's order count where its feed publishes one, at the depth of the recorded feed, every change in order. Per-order ids, queue positions and Level-3 (market-by-order) feeds, where a venue offers one, are not part of the archive; queue-position models work from the level sizes and the trade tape instead.

How many exchanges have Level-2 order book history in the archive?

31 venues recorded activity in the 30 days to 2026-11-17, every one of them with the Level-2 book next to $5.3T of traded turnover. Coinbase, Kraken Spot, BitMEX, Polymarket, Kalshi, predict.fun publish depth without a separate best-bid/ask stream, so top of book is derived by replay there; every other venue carries the BBO stream in the same file.