Recording 7 venues
Recording nowDaily files

Coinbase historical order book data

Kalshi Eye records Coinbase's public market data around the clock: every change to the order book and every trade, for 19 coins against USD. Each UTC day is published as Parquet files, one per coin. Coinbase sends its whole book, so every level is there.

Rows on file
332.7Mbook changes and trades
Coins with pages
19one per coin
Size
2.34 GBall datasets on file

Biggest Coinbase coins

All 19 coins

SeriesMarketsRowsSize
BTC/USD on Coinbasecb:BTC3 markets67.2M rows433 MB
ETH/USD on Coinbasecb:ETH3 markets59.6M rows382 MB
ZEC/USD on Coinbasecb:ZEC3 markets46.8M rows292 MB
XRP/USD on Coinbasecb:XRP3 markets17.1M rows156 MB
NEAR/USD on Coinbasecb:NEAR3 markets15.6M rows104 MB
LTC/USD on Coinbasecb:LTC3 markets15.3M rows96.8 MB
SOL/USD on Coinbasecb:SOL3 markets15M rows141 MB
ADA/USD on Coinbasecb:ADA3 markets13.6M rows109 MB
19 series on Coinbase with daily filesAll Coinbase series →

See the rows

First rows of each dataset's sample file.

Every change to Coinbase's whole order book for 19 coins, with a full snapshot each hour and after each reconnect.Full bookParquet·CSV

Datasets

One file per dataset per UTC day. Data dictionary

DatasetWhat it holdsPlanSample
Coinbase order book (L2, full depth)
coinbase_book
Every change to Coinbase's whole order book for 19 coins, with a full snapshot each hour and after each reconnect.
12 columns
  • ts_recvtimestamp[ns, UTC]
  • ts_venuetimestamp[ns, UTC]
  • assetstring
  • kindstring
  • sidestring
  • pricefloat64
  • sizefloat64
  • seqint64
  • checksumint64
  • price_dpint8
  • qty_dpint8
  • sessionstring
Full bookSampleParquet·CSV
Coinbase trades
coinbase_trades
Every Coinbase trade on the same coins, with the taker's side.
8 columns
  • ts_recvtimestamp[ns, UTC]
  • ts_venuetimestamp[ns, UTC]
  • assetstring
  • sidestring
  • pricefloat64
  • sizefloat64
  • trade_idstring
  • sessionstring
ResearchSampleParquet·CSV

What each plan opens

Founding members pay half, locked for life. Pricing

DatasetExplorerResearchFull book
Coinbase order book (L2, full depth)SampleSample✓
Coinbase tradesSample✓✓

Reading the files

  • One file per coin per UTC day. Replay in file order: a snapshot (or the hourly book copy) replaces the whole book, and each update sets one level's new size; 0 removes it.
  • A gap row means the book is unknown from that row until the next snapshot; its side says why.
  • Trades carry the taker's side: buy lifted the ask, sell hit the bid.
  • Timestamps are nanoseconds UTC: ts_recv when we received the message, ts_venue the exchange's time. Every dataset shares the recorder clock, so you can join to prediction market books on ts_recv.

Coinbase's API gives the book as it is now and recent trades; it keeps no past order books. Replaying depth, spreads or queue position needs a recording of the book as it changed, which is what these files are.

-- Every BTC trade on Coinbase for one day, in DuckDB
SELECT ts_recv, side, price, size, trade_id
FROM 'kalshieye-data/coinbase_trades/date=2026-11-12/asset=BTC/coinbase_trades-BTC-2026-11-12.parquet'
ORDER BY ts_recv;

Questions

How far back does the Coinbase data go?

Recording started on November 11, 2026 (a partial first day). Every day since then is in the archive.

Which Coinbase coins are recorded?

19 coins against USD: every change to the order book and every trade.

Can I try the Coinbase data for free?

Yes. Every dataset has a sample file, no account needed, linked in the table above.

Is Kalshi Eye affiliated with Coinbase?

No. Kalshi Eye is independent and not affiliated with Coinbase. We record Coinbase's public market data ourselves.