Online blackjack
how it really works.
Crypto-casino blackjack splits into two surfaces: in-house RNG variants (Stake Originals, BC Originals, etc.) and live-dealer studios (Evolution, Pragmatic Live, Ezugi). Both are fair under their respective verification regimes. Both kill card counting. The rule differences that actually matter for your edge are below.
What you are actually clicking on
Operator Originals
Stake Originals Blackjack, BC Originals, Roobet in-house variants. Software-driven; the operator generates virtual cards via RNG, paying out per the operator's posted rules. Provably-fair seed schemes let you verify individual rounds via SHA-256 hashes after the fact.
- • 60-150+ hands per hour, solo play
- • Fresh virtual deck per hand (no persistence)
- • Provably-fair seeds, verifiable post-round
- • Lower min bets ($0.10 typical)
- • Often 6:5 by default — verify before depositing
Evolution / Pragmatic / Ezugi
Cards dealt by a real human at a real table, broadcast over webcam. The crypto casino is reselling the studio's feed, not running the table. Most regulated-jurisdiction live blackjack online runs through one of three studios; the operator brand on the page is essentially the affiliate skin.
- • 40-80 hands per hour, dealer-paced
- • Multi-player tables, social atmosphere
- • Audited by eCOGRA / GLI (third-party)
- • Higher min/max bets ($1-$10,000+)
- • Standard 3:2 payout on most variants
Every dimension compared
| Dimension | RNG | Live dealer | What it means for you |
|---|---|---|---|
| How cards are dealt | Software draws from a virtual shoe; visuals are decorative. | Real human dealer at a real table, broadcast over webcam. | Cosmetic. RNG runs faster (60+ hands/hr) but is less social. |
| Fairness verification | Provably-fair seeds: client + server seed combine to a hash you can verify after the fact. | Independent test-house audits (eCOGRA, GLI) and continuous live observation. | Both are robust at scale. Provably-fair lets you check individual rounds; live audits cover the table. |
| Shoe state across hands | Reshuffled every hand. No persistence. | Continuous-shuffle machines or shoe-shuffle machines; persistence is minimal even when present. | Counting is impossible on RNG and almost-impossible on live. Discussed below. |
| Hands per hour | 60-150+ depending on the player. | 40-80 typical. | More hands = more turnover = more total expected loss at any given edge. Bet size matters more than session length. |
| Table limits | $0.10–$2,500 typical. | $1–$10,000+ depending on operator. | RNG is friendlier to bankroll exploration; live is required for $1k+ stakes. |
| Social element | Solo. Maybe a chat box. | Multi-player table feel, dealer interaction, table chat. | Pure preference. No EV implication. |
| Bet behind / multi-seat | N/A. | Common: bet behind another player's hand or play multiple seats simultaneously. | Bet-behind is functionally identical to playing your own hand at the same stakes. |
| Variance per session | Normal distribution; high hand count compresses variance over time. | Same EV math, lower hand count means individual sessions are more variable. | Same long-run math. Short sessions feel different on live; long-run identical. |
What the seed actually proves
Crypto casinos popularized provably-fair RNG: a cryptographic scheme that lets you verify, after the fact, that the operator didn't alter the outcome of a specific hand. The mechanic:
1. Operator generates server seed for the round. Hashes it (SHA-256), publishes the hash before play. You see: 7c1f5d… (the hash) You don't see: the underlying server seed. 2. You set a client seed (any string you want). The combination of server seed + client seed + nonce (round number) deterministically generates the cards via HMAC-SHA-512. 3. After the round, the server reveals the seed. You verify the hash matches what was published pre-round, then re-derive the cards yourself. If the math reproduces the cards you saw, the operator couldn't have peeked at your hand mid-round and tilted the result against you.
What it proves: the operator committed to the round outcome before play started and didn't change it during play. What it doesn't prove: that the seed wasn't generated to be unfavorable in the first place — though composing across many rounds, the statistics make sustained skewing detectable.
Provably-fair is a strong fairness primitive but is not a substitute for licensed-jurisdiction oversight. Most of the largest crypto casinos run both: provably-fair seeds for in-house RNG, plus jurisdictional licensing (Curaçao, Anjouan, MGA) for the operator-level accountability that catches sustained statistical deviation.
What the studios actually do
Evolution Gaming, Pragmatic Live, and Ezugi run the three largest live-dealer studios used by crypto casinos. Tables are physical: a real dealer, a real shoe, real cards, real turn-and-flip animations broadcast at 30-60fps to your browser.
Most live-dealer blackjack tables run 8-deck shoes (some 6-deck) with continuous-shuffle machines (CSMs) or shoe-shuffle machines (SSMs) doing the cycle. CSMs eat the discards into the shoe between hands; SSMs shuffle the entire shoe each cycle. Both eliminate shoe-state persistence, which is the only state that card counting depends on.
Audit regimes for live-dealer studios are mature: eCOGRA, GLI (Gaming Laboratories International), and in-jurisdiction regulators (UKGC, MGA) test studios regularly for RTP compliance, dealer training, and stream integrity. Cheating at studio scale is extraordinarily hard to hide; the surface area for fraud is too large and the regulatory oversight too continuous.
That said, the operator running the affiliate skin in front of the studio matters. Crypto operators on weak licences (Anjouan, some Curaçao sub-licences) have less independent recourse if a payout dispute arises with the studio feed. See /casinos for our trust-score breakdown by operator.
Why neither surface lets you count
Card counting requires shoe-state persistence: the cards that have been dealt and not returned to the shoe, so the remaining shoe's composition can drift toward 10-rich (favoring player). Both online surfaces specifically eliminate persistence:
- RNG: fresh virtual deck per hand. The shoe doesn't exist between rounds; what existed in round N is gone in round N+1. There is nothing to count.
- Live (CSM): discards feed back into the shoe in real time. The composition you'd be counting is being re-randomized while you count.
- Live (SSM): shoes reshuffle far before reaching the deep penetration counters need. Online live tables typically reach 20-30% penetration before the next shuffle, vs the 75% needed for a meaningful spread.
The full primer on counting + why it doesn't work online is on /games/blackjack/card-counting. The short version: the math is sound; the application is dead.
The rules that actually move the math
Since counting is off the table for both surfaces, the rules of the specific table you sit at are the only lever you have. In rough order of impact:
- Blackjack payout (3:2 vs 6:5) — +1.39% house edge for 6:5. The single most consequential rule. Walk past 6:5 even when the game is otherwise identical to a 3:2 table next door.
- Soft 17 rule (S17 vs H17) — +0.20% for H17. Dealer hits soft 17 means more dealer 18-21 hands but also more busts; net is player-negative.
- Double after split (DAS vs no-DAS) — +0.14% for no-DAS. Some pair splits stop being +EV without the option to double the post-split hand.
- Surrender availability — +0.08% for no-surrender. Worst hands lose the half-bet escape. Standard at most online live tables; rare at RNG.
- Number of decks — single-deck is best, 8-deck is worst. Total spread: ~0.50%. But almost-always paired with other rule changes that more than offset.
Full impact ranking on /games/blackjack/odds. The variant-aware chart on /games/blackjack/strategy lets you switch between rule sets via the chip strip or by clicking a casino logo.
What to actually do
Both RNG and live-dealer blackjack are fair under their verification regimes. Pick whichever matches your preference for pace, social atmosphere, and bet sizing. Then make sure the rules of the specific table are not ridiculous (3:2 payout is non-negotiable).
If you're betting under $10/hand, RNG is faster and friendlier to small-stakes play. If you're betting $50+/hand, live tables run a smoother game with better limits and lower variance per session. The long-run EV is identical between the two surfaces under matched rules.
The two surfaces also coexist on most operators. Stake, BC.Game, and the rest run both their in-house RNG Originals and a separate set of live-dealer tables from Evolution / Pragmatic Live. The rules can differ — one operator's RNG might be 6:5 while their live tables are 3:2, or vice versa. Verify the payout before sitting.