How to get a crypto implied volatility surface

The short way

Ask the venue. Nearly all of the listed BTC and ETH options market sits on a handful of exchanges, and the largest of them publishes its own marks to anyone who asks, with no key, no account and no plan.

GET https://www.deribit.com/api/v2/public/get_book_summary_by_currency?currency=BTC&kind=option
GET https://www.deribit.com/api/v2/public/ticker?instrument_name=BTC-<EXPIRY>-<STRIKE>-C

The first returns every listed option in that currency with mark_iv and underlying_price on each row — a whole surface, in one unauthenticated call, as a grid of strikes and expiries rather than as a fitted sheet. The second adds the rest of what an options desk reads off a single instrument: bid_iv, ask_iv, a greeks object carrying delta, gamma, vega, theta and rho, and interest_rate, documented as the "interest rate used in implied volatility calculations".

That last field is the whole of this page in one line. The exchange is not reporting a measurement; it is telling you which assumption it inverted its own prices against. Everything below is about who else makes that assumption, what they charge for it, and why their answer differs from this one.

What the options are

One venue's marks, free. The endpoints above, polled. Use them when the question is about the instruments that exchange lists and margins — a position's greeks, a strike's mark IV, the shape of the front expiry. You get no history at all, no cross-venue view, and no fitted values between the strikes that exist.

A screen rather than a feed. SignalPlus is the free one: a volatility lab plotting term structure, the smile, the surface, implied against realised, a volatility cone and time-lapse IV snapshots, over Deribit, Bybit, OKX, Binance and Coincall. Its homepage calls two of those displays the "Model" smile and the "Model" surface, which implies a fit rather than a pass-through — and the documentation never says whose fit. Laevitas is the paid screen, at 50 dollars a month for a year of history against one week on its free plan, with a multi-leg builder and an options backtester attached. Neither is a data feed; both are things you look at.

Fixed tenors, cheaply. Velo sells the shape rather than the sheet: DVOL as the volatility index, Black-Scholes greeks, ATM implied volatility and 25-delta skew interpolated to fixed 1-week, 1-month, 3-month and 6-month tenors, with realised volatility beside them. The coverage is Deribit BTC and ETH and nothing else. The terminal and an API key are one 199-dollar subscription, or two cents a call pay-as-you-go — and on monthly billing the history stops at three months, which is the detail that decides whether the price is the price.

A calibrated surface as an API. Laevitas is the card built for this job: per-instrument implied volatility and greeks, options trade flow, dealer gamma exposure, and a proprietary calibrated surface queryable as snapshots, slices, term structure or strike diagnostics — with inverse and linear margin books kept deliberately separate, which is the detail that tells you the surface is calibrated rather than scraped. A key is issued against a 500-dollar-a-month seat; the alternative is 0.10 USDC per request through x402, quoted in the payment challenge, with no key at all.

Surfaces on the axis you asked for. Amberdata sells delta surfaces and moneyness surfaces, each in a constant-maturity form for fixed horizons and a floating form for the venue's actual expiries, plus implied against realised and variance risk premium. It is the one here that writes the construction down: the mark implied volatilities at the neighbouring deltas are converted into variances, linearly interpolated, and converted back into implied volatility. Frequency is not uniform — minutely for Binance and for Deribit after September 2021, hourly for older Deribit, daily for OKX history. It is also the only card on this page with a checkout: 600 dollars a month buys one exchange in one market, and Deribit options is one of the 51 combinations sold that way, with the key provisioned in 24 to 48 business hours rather than at purchase.

Minute-granularity history, on a contract. Kaiko dates its implied-volatility surfaces and smiles to April 2021 at one-minute granularity, with derivatives contract details from July 2020 behind them. Nothing about it is self-serve: no published price, no free tier, and a key issued by the vendor's team against an active subscription. Why half this market will not quote a price is the background to that.

The chains, to fit your own. Tardis.dev sells an options_chain CSV carrying bid_iv, ask_iv, mark_iv, delta, gamma, vega, theta, rho, underlying_price, the strike, the expiry and open interest — and the schema is explicit that these come from the venue and are empty where the exchange does not provide one. Nothing in that file is computed by the vendor, which is exactly why you would buy it. History starts 30 March 2019 for the original venues, arrives the next day at around 06:00 UTC, and the Options plan starts at 350 dollars a month on the academic tier while the replay API starts at 1,000.

Where this breaks

A surface is a model over quotes, not an observation. Implied volatility is not measured; it is what you get by inverting an option's price through a pricing model, and every input to that inversion is somebody's choice. Which quotes are admitted and which are dropped as stale or too wide. What forward the strike is measured against. What rate goes into the inversion — Deribit publishes its own as a field on the ticker, and a vendor fitting the same instrument uses its own. And how the gaps are filled: Amberdata's documented method converts the mark IVs at the neighbouring deltas into variances, interpolates linearly, and converts back, which is a specific decision with a specific bias, and a vendor interpolating in volatility instead lands somewhere else from identical inputs. Two vendors' IV for the same instrument differing is the normal case, not a data-quality incident. If you are going to compare them, compare constructions first.

Mark IV and mid IV are different numbers, and only one of them margins your account. The exchange's mark is the price it uses for margin and liquidation, so mark IV is a published, load-bearing figure that exists whether or not anyone is quoting. An IV computed from mid is the volatility implied by a midpoint — and on an untraded strike the bid and the ask can sit far apart or not exist at all, which is why Deribit's book summary documents bid_price and ask_price as null when there are no bids or asks. Three separate fields, three separate claims. A vendor that does not say which one it is showing you has not told you what you bought, and SignalPlus, for all that its vol lab is the best free one here, is the clearest case: neither its volatility lab page nor its analysis widgets say whether the greeks are its own calibration or the venue's marks.

The grid is sparse, and it gets sparser away from the front. One underlying carries hundreds of strike-and-expiry combinations and only a handful trade on any given day. Interpolation between two live points is one thing; extrapolation past the last one is another, and the honest vendors stop rather than guess. Amberdata's delta surface returns null for the target when the outside delta point it needs does not exist. Tardis.dev's columns come back empty where the venue published nothing. A far-dated 25-delta point is frequently a value derived from two quotes that are themselves wide, stale, or the market maker's placeholder — and it will still render as a smooth line on a chart, which is the part that misleads.

Every filter is a judgement, and index methodologies make that visible. DVOL is worth reading even if you never use it, because Deribit publishes what it throws away: option prices are taken as volume-weighted averages walking the order book until 2 BTC of cumulative depth, in-the-money calls and puts are discarded along with far out-of-the-money options below 5% delta, instruments listed less than an hour earlier are excluded, the two expiries either side of 30 days are interpolated through variance-swap arithmetic, and the result is smoothed with an exponential moving average over 240 points. Those are all defensible. They are also all decisions, and a vendor's surface that publishes none of them has made the same kind of decisions silently.

A market surface is mostly one venue's book. Every vendor on this page that names its options venues names the same short list. Laevitas's public instrument catalogue returns chains for exactly four — Deribit, OKX, Binance and Bybit — against a marketing counter that says 15+. SignalPlus publishes dashboards for five and key guides for a sixth, which is close to the whole tradable universe. Amberdata's coverage table carries eight options venues out of 65 exchange-and-market rows. So "the market's implied volatility" is a weighted view of a handful of books, dominated by whichever of them has quotes on the strike you asked about, and the weights are rarely published. The options shelf is the catalogue read this way, and where the Bitcoin price comes from is the same problem one layer down, on the underlying every one of these surfaces is struck against.

Margin flavour splits the chain in two. A BTC-margined inverse option and a USDC-margined linear option on the same underlying, strike and expiry are two instruments with two books, two marks and two implied volatilities, and merging them produces a surface that describes neither. Laevitas keeps the inverse and linear books separate for this reason. If your loader normalises by strike and expiry alone, it is silently averaging two markets.

Series end. Laevitas publishes a last observation for each of eight retired options venues — Zeta in September 2022, Lyra and Lyra on Arbitrum in December 2023, PowerTrade in March 2024, AEVO in June 2024, BIT and Delta Exchange in April 2025, CME options in December 2025. A backtest that spans those dates is a backtest whose universe changes underneath it, and nothing in a surface endpoint will tell you that the venue stopped rather than that volatility went quiet.

And several products that look adjacent do not compute this at all. CoinGlass covers options on four venues with open interest, volume and max pain and says plainly that it has no implied volatility, no greeks and no chain; Coinalyze has no options anywhere; Paradigm is execution and computes nothing, reading delta off the venue's own market data for a hedge leg rather than fitting it. Positioning data — open interest, max pain, the options-to-futures ratio — is a different purchase from a volatility surface, and it is the substitution most likely to be made by accident.

If you outgrow this

If the problem is that the number has to be reproducible — it is going into a mark, a risk report or somebody else's screen — stop buying surfaces and buy chains. Tardis.dev's files carry the venue's own IV and greeks alongside the bid and the ask, so you can refit with your own forward, your own rate and your own filters, and write the method down. Read the licence first: redistribution is barred except at ten-minute resolution or coarser, and the machine-learning clause is stricter than most.

If the problem is history depth, that is the axis the money is on. Kaiko is the deepest fitted series here at one-minute granularity from April 2021; Amberdata's minutely frequency only applies to some venues and cadences, with older Deribit hourly and OKX history daily; and Velo's monthly billing opens three months where its yearly opens the archive. Check the granularity per venue before assuming a uniform grid, because none of these are uniform.

If the problem is that you want to look rather than pull, the screens are an order of magnitude cheaper than the feeds. Laevitas Platform is 50 dollars a month against 500 for the key, and SignalPlus costs nothing — but note what SignalPlus is: a terminal with an order ticket that holds your exchange keys on its servers, which is a different trade from a data subscription. Paradigm against SignalPlus is where that distinction is worked through.

If the problem is cost per call rather than cost per month, Laevitas's x402 path at 0.10 USDC a request and Velo's two cents a call are the two metered options in this corner of the catalogue, and both are cheap for a lookup and expensive for a backfill. An agent left looping against either is a bill. They differ on what you can plan around: Velo publishes a rate limit of 120 requests per 30 seconds and caps a single response at 22,500 values, while Laevitas publishes no REST limit anywhere and returns your ceiling in response headers, which makes capacity a support conversation.

And if what you actually want is a price to trade a structure at, no surface on this page is that. A fitted IV tells you what a model says an option is worth; what you can get filled at on a multi-leg structure comes from an RFQ, which the trading terminals category is where you would look for, and a different kind of account.