Every open perpetual bet we can see, venue by venue, counted ONE side only and ONE principal instrument per venue. Steps in the total are composition — venues joining or falling out of a day — not the market moving; the «how many venues» view prints the count.
$31.81B BTC · venue by venue, in dollars · all 18 venues added up · $ notional on Oct-22 2026, the last day in this build, read from the venues' records, published by askthetape.com.
An open bet needs two people: one betting the price rises, one betting it falls. Exchanges report the pair in different ways — some count both sides, some count one — so adding their raw numbers double-counts half the market. Here every venue is brought to ONE side before it is added, and the page says for each one HOW that side is known. Only 9 exchanges publish a history of this at all, which is why 6 of the 18 lines begin on 2026-09-11: that is the day our own five-minute capture started, and for the rest no public archive exists to fill the years before it. On 2026-10-22, the last closed day, the 18 venues held $31.8B of bitcoin bets between them — Binance USD-M 24.5% · MEXC 11.8% · Bitget 9.4%. How much of that total is proven and how much is taken on trust, by dollars: 62.4% is taken at the field's own name, with nothing contradicting it · 16.7% is matched against the Open Interest the exchange shows its own traders in the same second · 15.4% is proven on a second handle of the exchange itself · 5.5% is both sides by the venue's own note, confirmed by our own record of its positions, and halved. In plain words, the biggest lines on the page are in the unverified class: Binance USD-M 24.5%, Bitget 9.4%, OKX 7.9%.
| leg | exchange public APIs (18 venues, one snapshot every five minutes, 188 requests a run) · own 5-minute capture since 2026-09-11 · own prod capture of Binance, Bybit, OKX, Bitget and Hyperliquid since 2026-03-31 · exchange archives: Bybit since 2020-08, Binance USD-M since 2020-09 (BTC only; the other coins start 2021-12-01, measured against the dump listing, not assumed), Binance COIN-M since 2021-07, Bitfinex since 2020-01, Kraken since 2023-03, OKX since 2023-12 |
|---|---|
| source of the Binance archive | Binance's public daily archives — the two Binance archive legs on this page (USD-M, whose contracts are margined in dollars, from 2020-09; COIN-M, coin-margined — the margin is the coin itself — from 2021-07) come from Binance's own published daily archive, not from a live endpoint and not from our capture. Every other venue's history is that exchange's own archive endpoint; the recent end of every line is our own five-minute capture. |
| calibration | one side per venue, and the page says for each venue HOW that side is known. Bybit and Gate publish both sides in their headline field, and this page does not halve that field: it takes the exchange's own ONE-SIDED handle instead (Bybit singleOpenInterest, Gate contracts.position_size). The two-sided field is ×2.000000 and ×2.0004 of what we carry — measured on the exchanges' own second handle on 2026-09-11, which is what proves the handle we read is one side. 🩸 Our own prod capture of Bybit is a different matter: it does carry both sides, so it IS halved — measured against Bybit's archive on 810 common venue-coin-days (median ratio 2.0000); Binance 1.0002 and OKX 1.00 are left alone. Classes of evidence: verified = proven on a second handle of the exchange itself (Bybit, Gate) · position-verified = both sides by the venue's own note, confirmed by our own record of its positions, and halved (Hyperliquid) · ui-verified = matched against the Open Interest the exchange shows its own traders in the same second (MEXC, WhiteBIT) · unverified = taken at the field's own name, with nothing contradicting it. 🩸 The most-quoted public aggregator carries Bybit and Gate at their two-sided fields — 61.53K and 63.98K BTC at 22:23:33Z on 2026-09-11, where the proven one side was 26.42K and 31.92K — so a total built here is not comparable with a total built there until that correction is made. 🩸 One venue, one instrument: each venue is carried by ONE principal instrument per coin — the USDT-settled one when exactly one exists, else the only instrument, else the largest by open interest in that snapshot (printed as «largest»); every other perpetual book of that venue is counted separately and never added into the line. On 2026-10-22 the rule picked usdt 220 · single 122 · largest 4, and «largest» is never silent — it is printed for Deribit BTC-PERPETUAL, Deribit ETH-PERPETUAL, Kraken PF_ETHUSD, Kraken PF_XBTUSD. 🩸 Hyperliquid reports two-sided open interest — its own note of 2026-09-23: «Two-sided OI is calculated as the sum of long and short positions» — and our own record of its positions agrees (on no market do the tracked longs reach much past half of the field); we halve it to one side, like the other venues, over its whole history from 2026-03-31. Until this correction the page counted it at both sides, so its share and each coin's total sat above one side. Aster and Paradex stay unverified: their docs do not name the side, and we hold no record of their positions. dYdX counts one side (its indexer sums open longs). |
| 🩸 dollars or coins — the unit changes the answer, and so does the venue set | The headline lines on this page are DOLLARS of open bets ($ notional), and a dollar total moves for two different reasons: somebody opened a bet, or the price moved under the bets that were already open. Read it LIKE-FOR-LIKE — on the 5 venues that report on both days, the same fixed set the like-for-like dollar line uses — and over the last thirty days (2026-09-22 → 2026-10-22) bitcoin's dollar total moved -10.5 %, the same bets counted in COINS moved -5.3 %, and the price itself -5.5 %. Both counts decreased over these dates: -10.5% in dollars and -5.3% in coins. 🩸 And the total over ALL venues is a third number again: the set itself grew from 19 venues to 18 over that window, its coin total moved -8.1 %, and 0.0 % of it on the last day comes from venues that were not in the set on the first day — that part is composition, not anyone opening a bet. Every coin group therefore carries a second view, «the same bets counted in coins, not dollars», whose bold line is the fixed set and whose thin grey line is the growing total; it is built from the exchanges' own coin figure in the very same row, nothing is divided by a price. The unit check is re-run on every build: dollars ÷ coins is the price, and on the last five closed days of the fixed set it reads 2026-10-18 $82,023 · 2026-10-19 $81,936 · 2026-10-20 $83,009 · 2026-10-21 $84,105 · 2026-10-22 $81,373. The «all coins» view stays in dollars only — adding SOL to BTC in coins would be adding apples to oranges. |
| cadence | every five minutes, last snapshot of the UTC day; the archives are the exchanges' own daily closes |
| since | per venue and coin, never one blanket date: Bybit BTC 2020-08-04, Binance USD-M BTC 2020-09-01, Binance COIN-M 2021-07-08, Bitfinex 2020-01-01, Kraken 2023-03-07, OKX 2023-12-31, Bitget and Hyperliquid 2026-03-31 (our prod capture), Gate 2023-10-11, dYdX 2023-11-13, Paradex 2025-11-15, HTX 2026-03-13 (archives found on 2026-09-14), the other 6 venues 2026-09-11 (our own capture, day 1) — 18 named, 18 in the total |
| 🩸 archives found 2026-09-14 | Four more venues turned out to publish a history of open interest after all, and they are spliced in here: Gate — api/v4 futures/usdt/contract_stats interval=1d, history since 2023-10-11: the stamp of day D is the CLOSE of day D. Our own five-minute capture over the days both cover reads 1.00001 for bitcoin, 0.99997 for ether, 1.00044 for solana — the median of the three closed days both legs cover, measured on 2026-09-14. The label itself is proved WITHOUT us: the exchange's own daily bar equals its own 23:00Z hourly bar to six decimals on 7 closed days out of 7. 🩸 its headline field counts BOTH sides, so it is divided by exactly 2 — that is what both sides means, not a number tuned to match our own capture. Read live, that ratio wobbles between 1.9976 and 2.0120 depending on which second each handle is read; a reading outside 1.9-2.1 would drop the leg rather than splice it quietly · HTX — linear-swap-api/v1/swap_his_open_interest period=1day amount_type=2, history since 2026-03-13: the stamp is 16:00Z — midnight in the exchange's own timezone — and the value is the reading at that instant, shifted onto day D by 8 hours, which we declare rather than hide. Our own five-minute capture over the days both cover reads 0.99953 for bitcoin, 0.99772 for ether, 0.99995 for solana against our 23:55Z close, and 1.000000 exactly when both are read at the 16:00Z stamp itself (2026-09-14). 🩸 the handle returns only the last 185 bars and takes no time cursor, so the window SLIDES: every day not taken is a day of archive lost for good, which is why the raw response of this pull is kept. Its last bar is 2026-09-13 16:00Z · dYdX — v4/candles/perpetualMarkets resolution=1DAY, field startingOpenInterest, history since 2023-11-13: the candle of day D carries the open interest at the START of D = the close of D−1, so the row is shifted back one day (the same convention as Bybit). Our own five-minute capture over the days both cover reads 1.00000 for bitcoin, 1.00001 for ether, 0.99997 for solana (2026-09-14), against 0.98931 if the dates are matched naively; and, again without us, the exchange's own daily bar equals its own 00:00Z hourly bar on 8 days out of 8. 🩸 the exchange publishes no dollar figure, so the dollars are the coins times that day's own close · Paradex — v1/markets/summary over a start/end window, history since 2025-11-15: five-second snapshots; the last one before midnight is the close of day D. Our own five-minute capture over the days both cover reads 1.00000 exactly on all three coins (2026-09-14). 🩸 the level is fifty times lower today than in January 2026 — judged before splicing, and the deciding number is the venue's own RUNNING TOTAL of traded volume, which no redenomination can move: it added $37.5B over the three months to March 2026 and only $2.76B over the six months after, a thirteenfold slowdown. Alongside it: one market and one unit throughout (listed 2023-10-04, size step 0.00001 BTC), open interest against 24-hour volume drifting across a 0.22-3.62 band with no break, and a largest single-day move of −43 % on 2026-03-26 on unchanged volume. An activity collapse, not a unit change. 🩸 the other 9 venues do not publish one: Aster · Bitget · Coinbase Intl · Crypto.com · Deribit · Hyperliquid · KuCoin · MEXC · WhiteBIT — 26 endpoints tried live on 2026-09-14, every one of them dead (404 / «method not found» / «V1 decommissioned»), so «no public history» here is a RECORDED attempt, not a shrug. Priority where two legs cover the same venue-day is unchanged: our own five-minute capture first, our prod capture second, the exchange's archive last — the archive sets the DEPTH, never overrides a day we measured ourselves. The backfill is a one-off: from 2026-09-11 these venues are carried forward by our own capture. |
| composition | 🩸 the total is a sum over the venues that EXIST that day, and its 12 composition steps — listed here, counted off the data, with the four most recent of them also marked on the chart — are composition, not the market. Every date a venue enters: 2020-08-04 Bybit · 2020-09-01 Binance USD-M · 2021-07-08 Binance COIN-M · 2023-03-07 Kraken · 2023-10-11 Gate · 2023-11-13 dYdX · 2023-12-31 OKX · 2025-11-15 Paradex · 2026-03-13 HTX · 2026-03-31 Bitget, Hyperliquid · 2026-09-11 Coinbase Intl, Crypto.com, WhiteBIT · 2026-09-12 Aster, Deribit, KuCoin, MEXC. 🩸 One step goes the other way and is NOT in that count, because the count is of entries: on 2026-09-08 the set fell 12 → 11 for a day, a venue we could not read that day rather than a venue that closed. Two entries carry most of the weight: 2026-03-31 (Bitget and Hyperliquid enter from our prod capture) and 2026-09-11 (6 more enter from our own capture). The «how many venues that day» view carries that number as its own line, so a step can always be told apart from a move. 🩸 A step is never ONLY new venues: on 2026-09-11 the 11 venues that existed on both days step by $-0.79B = -3.2% of that day's total, while the venues joining bring $3.33B = 13.3%. 🩸 That first part is no longer a change of definition: each venue is carried by ONE principal instrument per coin — the USDT-settled one when exactly one exists, else the only instrument, else the largest by open interest in that snapshot (printed as «largest»); every other perpetual book of that venue is counted separately and never added into the line — the same contract the exchange archives and our prod capture measure. On 2026-10-22 the 18 venues our own capture reaches held $31.810B of bitcoin in principal contracts and $33.554B counting every perpetual book of those venues; that difference is printed here and never added into the line. The Bitget prod leg matches our own capture to within 0.00% on the 172 days where both existed (median ratio 1.0000). 🩸 62 own-capture rows (2026-09-11) could not name their principal instrument: the capture of that day still recorded a venue's coin as ONE lump of every book it runs (Bybit 18 · MEXC 18 · Binance USD-M 9 · OKX 8 · Aster 3 · KuCoin 3 · Deribit 2 · Bitfinex 1), and a lump cannot say which book is the principal one — so those rows are not published. 15 of them are still carried that day by another leg (our prod capture 9 · the exchange archive 6); 47 of those venue-coin-days are a HOLE, not a guess, and the «how many venues» line steps down with them until the per-instrument capture covers the day. Shares are therefore printed only INSIDE one composition epoch and never across a seam. |
| 🩸 a frozen handle | no venue is printing a repeated open interest in the last thirty days of this build. 🩸 62 archive rows (Binance USD-M 20 · Bybit 21 · OKX 21) were left out for the opposite reason: an archive pulled mid-day asks a T+0 handle (openInterestHist period=1d) that answers with YESTERDAY's bucket, so its last date is an echo, not a print of that day — and taking it for a print made the frozen-handle guard drop the honest day before it. |
| 🩸 one venue, two handles | Binance USD-M is on «Open bets and who leans where» too, and the two pages do NOT print the same dollars: here it is our own five-minute capture of fapi/v1/openInterest (BTCUSDT) times the mark price, last snapshot of the UTC day; there it is Binance's own hourly openInterestHist.sumOpenInterestValue, last hourly print of the day. On 2026-10-22 that is $7.8011B here against $7.7877B there, +0.17% apart — same venue, same day, two handles and two stamps of «end of day». Neither is rounded to hide the other. |
| 🩸 perp DEXs' share — what is counted, what is not proven | Measured: the part of each coin's perpetual open interest in coins (one principal instrument per venue) held on Hyperliquid (its main venue, no HIP-3 markets), Aster, dYdX and Paradex, with the same set of venues on top and bottom of the fraction each day. Oct-22 2026, as reported / if Aster and Paradex count both sides: BTC 7.06% / 6.36% (18 venues) · ETH 8.36% / 7.83% (18 venues) · SOL 11.52% / 9.69% (18 venues). In coins on that day: BTC: Hyperliquid 21,494 · DEXs 27,611 of 390,931 · ETH: Hyperliquid 545,987 · DEXs 642,596 of 7,689,039 · SOL: Hyperliquid 2,742,910 · DEXs 4,320,983 of 37,494,434. A venue joins the set on its first day in the data — BTC/ETH dYdX Nov-13 2023, Paradex Nov-15 2025, Hyperliquid Mar-31 2026, Aster Sep-12 2026 · SOL dYdX Nov-14 2023, Paradex Nov-15 2025, Hyperliquid Mar-31 2026, Aster Sep-12 2026; Hyperliquid's line starts on Mar-31 2026 because our record of it starts there, and Aster's starts at full size on day one (BTC 1.52%, ETH 1.44%, SOL 4.23%): steps of the set, not bets arriving. 🩸 Side, checked 2026-09-26: dYdX counts one side (its indexer sums open longs). Hyperliquid counts both sides — its own note of 2026-09-23: «Two-sided OI is calculated as the sum of long and short positions» — and our own position record agrees: over 178 markets tracked longs reach at most 0.517 of it and longs plus shorts at most 1.028 (median 0.775) — a ceiling at one, not two; so its line is one side over its whole history, from the same leg as the rest of the page. Aster's docs do not describe the field; Paradex's do not name the side. Not measured: volume, liquidity, bets held for reward points; not a forecast. |
| policy | derived aggregate by askthetape.com · no export. Option notional is never added to perpetuals: every row carries its market (perp / future / option) and only perps are summed — Deribit's bitcoin option board alone is about the size of every perpetual on earth. · read on the site · not exported: this page is drawn from someone else's archive |
| gaps | measured on this leg, not assumed: the line has no point on 2021-10-05 … 2021-10-05 (1 day) on Binance COIN-M; 2021-10-08 … 2021-10-08 (1 day) on Binance COIN-M; 2022-06-08 … 2022-06-08 (1 day) on Binance COIN-M — it BREAKS there and nothing is drawn across it |
| the same venues in coins, back to 2020 | for a long, clean history in COINS (not dollars) on fixed venues see «Bitcoin open interest, two venues since 2020», «Ether open interest, three venues» and «Solana open interest, two venues»: no composition steps there, a sum and a share view across the whole history. This page is the wide, dollar-denominated view of who holds the bets today, venue by venue. |
| hourly window | our own 5-minute capture oi_agg, last snapshot of each hour · since 2026-09-11 17:00Z by data, 1,012 hours with data inside a 1,013-hour span → 2026-10-23 21:00Z (the Tape and Desk windows draw up to 400 days of that, never more than we hold) · captured 2026-10-23T21:21:56Z |
| this build | data to Oct-22 2026 |
| policy | no backfill · no export · read it here · read on the site · not exported: this page is drawn from someone else's archive |
The chart itself is drawn in the browser: turn JavaScript on to hover a day, pick a window and draw your line. The numbers above come from the last build, Oct-23 2026 21:24Z.
What the tape says today · what's new · all 111 metrics