Choose the next move: Hit 🃏 or Stand ✋
Hit
Stand
BlackjackPilot Blog
We tested Hi-Lo, KO, REKO, Red Seven, Hi-Opt I, Hi-Opt II, Omega II, Zen Count, and Wong Halves across 5.4 billion simulated blackjack hands to rank each system by SCORE, EV, risk, and complexity.
CardCountingBenchmarkOverallTable, CardCountingBenchmarkScenarioTable, CardCountingBenchmarkScoreChart, } from '../CardCountingBenchmarkTables'
CardCountingBenchmarkCompareLauncher, CardCountingBenchmarkWinnerCompareLinks, } from '../CardCountingBenchmarkCompareLauncher'
Card-counting systems are usually compared from several different angles: tag values, betting correlation, playing efficiency, ease of use, and performance in a favorite rule set. BlackjackPilot's Card Counting Systems Compared guide covers the first four for all nine systems in this test; this benchmark adds a practical fifth:
If several counting systems use the same 1-10 ramp policy, with system-specific edge-calibrated ramps, the same bankroll assumption, the same seed list, and the same rule sets, which ones actually produce the best risk-adjusted results?
This benchmark compares nine blackjack counting systems across six common game conditions using the same engine, same seed matrix, and the same edge-calibrated betting policy. If you want the broader simulator validation context, see how BlackjackPilot validates its accuracy.
In this run, Wong Halves finished first overall, with Zen Count and REKO close behind.
The ranking below is by point estimate. When systems are close, treat them as a statistical cluster rather than a permanent order of superiority.
The top three systems are separated by about one SCORE point on average (72.69 vs 71.60). That gap is smaller than what you usually get from better game selection, deeper penetration, or a stronger spread.
That ranking should not be read as "Halves is always the best system for every player." It is the best point estimate in this benchmark. The top group is close enough that training cost, error rate, casino conditions, and how well you can actually execute the system matter as much as the raw table.
My practical read:
The best theoretical system is not always the best system to learn.
| Player type | Best fit | Why |
|---|---|---|
| New counter | Hi-Lo | Best documentation, easy training path, strong baseline |
| Wants no true-count conversion | REKO | Very competitive result with running-count play |
| Wants maximum benchmark performance | Wong Halves | Best average SCORE in this test |
| Wants strong but less extreme complexity | Zen Count | Top-tier result without fractional tags |
| Already trained on Hi-Lo | Stay with Hi-Lo or move to Zen | Hi-Lo remains strong; Zen is a logical upgrade |
| Wants an ace-neutral system | Hi-Opt II | Better result than Hi-Opt I in this setup |
The benchmark includes nine systems:
| System | Balanced? | Count Axis | Difficulty | Notes |
|---|---|---|---|---|
| Hi-Lo | Yes | True Count | 5/10 | Standard reference system |
| KO | No | Running Count | 5/10 | Unbalanced, no true-count conversion |
| REKO | No | Running Count | 5/10 | Simplified KO-family system |
| Red Seven | No | Running Count | 5/10 | Unbalanced system with red sevens counted differently |
| Hi-Opt I | Yes | True Count | 4/10 | Ace-neutral, simpler than Hi-Opt II |
| Hi-Opt II | Yes | True Count | 6/10 | Stronger ace-neutral system |
| Zen Count | Yes | True Count | 7/10 | Level-two balanced system |
| Omega II | Yes | True Count | 7/10 | Level-two balanced system |
| Wong Halves | Yes | True Count | 8.5/10 | Fractional-tag system with high betting correlation |
Difficulty scores are derived from each system's tag vector in BlackjackPilot (tag complexity, balance, and related ease metrics), scaled to a 1–10 scale where lower is easier.
Each system was tested in six scenarios:
| Scenario | Decks | Rules | Penetration |
|---|---|---|---|
| Single Deck | 1 | H17, no DAS, no surrender | 75% |
| Double Deck | 2 | H17, DAS, no surrender | 75% |
| Six Deck LS | 6 | S17, DAS, late surrender | 75% |
| Eight Deck | 8 | H17, DAS, no surrender | 75% |
| Six Deck H17 | 6 | H17, DAS, no surrender | 75% |
| Six Deck Deep | 6 | S17, DAS, no surrender | 85% |
Every system-scenario pair used:
Every system used the same betting shape: minimum 1 unit, maximum 10 units. The difference is how those bet sizes were assigned to count buckets.
For each system, BlackjackPilot built a native edge map: at each true-count or running-count bucket, it estimated that system's player edge from simulation. The ramp then mapped edge to bet size on the same 1-10 scale for all nine systems. A +2 true count in Hi-Lo and a +2 true count in Zen Count do not imply the same edge, so they do not get the same bet size — but the ramp policy itself is shared.
That keeps betting comparisons fair: systems are not rewarded for using a hotter native ramp that happens to fit their tags better.
The benchmark used BlackjackPilot's simulateHandsV51 engine and generated native edge maps for each count system. Balanced systems used true-count buckets. Unbalanced systems used running-count buckets with their correct initial running counts. The bankroll and risk metrics follow the same general ideas covered in the bet spread and Risk of Ruin playbook.
The article tables are built from the benchmark summary export, not from hand-by-hand traces. For transparency, the summary CSV is available here:
That CSV includes the system, scenario, SCORE, EV, player edge, risk metrics, count axis, ramp policy, index tier, rules, seeds, total hands, and confidence interval fields used in the sortable tables above.
You can generate similar article-level data from the app UI: open the Blackjack Simulator, switch to Custom Strategy, match the rules, counting system, index deviations, bet ramp, bankroll assumptions, and run size, then export the summary CSV from the results panel. The UI is built for smaller reproducible runs than this full 5.4B-hand benchmark, but it lets you test the same kind of EV, SCORE, Risk of Ruin, RTP, and per-hand diagnostic data for your own strategy setup.
The main ranking metric is SCORE.
SCORE is useful because it combines expected value and volatility into one risk-adjusted number. A system can show a slightly higher player edge while still ranking lower if it needs a noisier bet pattern to get there.
That matters in this benchmark. For example, REKO's average player edge is slightly higher than Zen Count's in the overall table, but Zen's volatility profile gives it a slightly higher average SCORE.
The headline result is that Wong Halves won five of six scenarios by point estimate. Zen Count won the single-deck scenario.
| Scenario | Winner | SCORE | Second Place | SCORE | Gap |
|---|---|---|---|---|---|
| Single Deck, H17, no DAS | Zen Count | 250.48 | REKO | 249.87 | 0.61 |
| Double Deck, H17, DAS | Wong Halves | 106.23 | Zen Count | 106.10 | 0.13 |
| 6D, S17, DAS, LS | Wong Halves | 31.25 | Hi-Lo | 29.47 | 1.77 |
| 8D, H17, DAS | Wong Halves | 3.90 | REKO | 3.57 | 0.33 |
| 6D, H17, DAS, 75% pen | Wong Halves | 12.23 | Hi-Opt II | 11.24 | 0.99 |
| 6D, S17, DAS, 85% pen | Wong Halves | 39.95 | Zen Count | 39.41 | 0.54 |
Important caveat: the first-place and second-place confidence intervals overlap in every scenario. So the right interpretation is not "Halves proved it always beats the field." The right interpretation is:
In this 5.4 billion-hand benchmark, Wong Halves had the best overall point estimate, but the top systems are close enough that execution quality can dominate the theoretical gap.
The 8-deck H17 DAS no-surrender scenario is the weakest game in the benchmark. Even the winner, Wong Halves, posted only a 3.90 SCORE with a 51.58% Risk of Ruin under the fixed 500 unit bankroll assumption.
That does not mean 8-deck blackjack is never playable. It means this specific play-all, no-surrender, 75% penetration setup is thin. Game selection, wonging, deeper penetration, a different spread, or a larger bankroll would matter more here than the small difference between most counting systems.
Wong Halves has excellent betting correlation. It is designed to tell you when the remaining shoe is favorable for larger bets. Since this benchmark uses the same 1-10 ramp policy with system-specific edge-calibrated ramps, that strength shows up directly in SCORE.
The cost is complexity.
Halves uses fractional tags. In practice, that means more mental overhead, more room for mistakes, and a slower training curve. A player who executes Halves with errors may perform worse than a player who executes Hi-Lo or REKO cleanly.
So the practical conclusion is:
Wong Halves is the best performer in this benchmark, but not automatically the best system to learn first.
REKO finished third overall with an average SCORE of 71.60, only behind Wong Halves and Zen Count.
That is notable because REKO is unbalanced and does not require true-count conversion. In this benchmark, the correct initial running counts were applied for each deck count, and the betting edge map was built in native running-count space.
That matters. If you treat an unbalanced system as if it starts from zero in every shoe, you will misread the count. Here, REKO's running-count axis was evaluated natively.
For players who care about practical execution, REKO's result is the most interesting part of the table. It did not win any scenario outright, but it was consistently strong enough to compete with more complex systems.
That is why REKO may be the practical winner for many players: it gives up very little in this benchmark while removing the true-count conversion step. A harder system only helps if the player can execute it accurately at real table speed.
Hi-Lo finished fifth overall:
| Metric | Hi-Lo Result |
|---|---|
| Avg SCORE | 68.46 |
| Avg Player Edge | 1.021% |
| Avg Scenario Rank | 4.50 |
| Difficulty | 5/10 |
That does not make Hi-Lo weak. It makes it a strong reference point.
Hi-Lo is easier to teach, easier to audit, and much more widely documented than most advanced systems. If you want a system with abundant training material and a good balance of power and simplicity, Hi-Lo still makes sense.
It also performed especially well in the 6-deck S17 DAS late-surrender scenario, finishing second behind Wong Halves.
Omega II and Hi-Opt I are not bad systems. Their lower ranking here is a result of this exact benchmark setup:
Ace-neutral systems can benefit from side-counting aces. This benchmark did not give Hi-Opt I or Hi-Opt II an ace side count, because the goal was a medium-complexity, apples-to-apples comparison rather than a maximum-performance advanced setup.
That is why I would not use this article to say "Omega II is bad" or "Hi-Opt I is not worth learning." I would say:
Under this simplified I18+Fab4 benchmark, Omega II and Hi-Opt I did not beat the top cluster.
Red Seven finished last overall.
| Metric | Red Seven Result |
|---|---|
| Avg SCORE | 52.81 |
| Avg Player Edge | 0.828% |
| Avg Scenario Rank | 9.00 |
| Avg Risk of Ruin | 35.36% |
This does not mean Red Seven is useless. It means this exact implementation, source index set, and 1-10 edge-calibrated play-all benchmark policy did not favor it.
Red Seven also has multiple published and practiced variants. A different index package, betting ramp, or wonging policy could move its result, so I would read this as a result for this benchmark implementation rather than a universal verdict on every Red Seven variant.
It also had the weakest 8-deck result, where all systems struggled but Red Seven's SCORE dropped to 0.67.
For beginners choosing between simple unbalanced systems, this benchmark points toward REKO or KO before Red Seven.
Click any column header to sort. Use the scenario buttons to narrow the table, and expand a row to see RTP, SD, confidence interval, win/loss/push rates, ramp metadata, index tier, surrender availability, and best count zones.
This is a practical benchmark, not a universal mathematical ranking of every possible implementation.
It does not test:
Those omissions matter. A more complex count can look stronger in a clean simulator than it does in a casino if the player makes more mistakes. A simpler count can win in practice because the player executes it faster and more accurately.
Late surrender also matters. This benchmark includes one late-surrender shoe scenario, but surrender rules vary by casino and can change the value of Fab4 deviations. For a rules-level explanation, see the blackjack surrender strategy guide.
The most useful way to read this table is in tiers.
Top cluster: Wong Halves, Zen Count, REKO, Hi-Opt II These systems produced the strongest overall results. Wong Halves had the best point estimate, but the top group is close.
Strong baseline: Hi-Lo, KO These systems remain practical and competitive. Hi-Lo is still the best standard teaching system; KO remains a strong running-count option.
Lower in this setup: Omega II, Hi-Opt I, Red Seven These systems may perform differently with side counts, full indices, or different betting assumptions, but they did not beat the top group here.
If you only care about this benchmark's point estimate, Wong Halves won.
If you care about practical execution, the story is more nuanced:
The biggest takeaway is not that every player should switch to Halves. The biggest takeaway is that system choice matters less than many people think once you compare strong systems under the same rules, same ramp policy, and same seed matrix.
For most players, a system you can execute perfectly is better than a theoretically stronger system you cannot execute under pressure.