You're playing Blackjack and the dealer has a face card showing. You have a 10 and a 2. If you stay, there's the possibility that the dealer has any card higher than a 2. But if you hit, there's the possibility you get a 10 or a face card. What's it going to be?

GPT-5.4

Vote: Hit

I maintain vote A: Hit. A hard 12 against a dealer 10-value upcard is a classic hit in basic strategy. Standing on 12 mostly concedes the hand, because the dealer will finish with 17-20 far more often than they will bust. The pro-stay argument is not persuasive because it misstates standard strategy and overweights the fear of busting. Yes, hitting can bust if a 10-value card comes, but standing on 12 against a face card is already a very poor position. You need to improve. Double down is clearly wrong on hard 12 versus such a strong dealer card. It magnifies losses on a weak holding. The strongest opposing point was the appeal to basic strategy from Sonar Pro, but its core claim conflicts with standard blackjack charts. The broader consensus here is also correct: hit is the percentage play.

Sonar Pro (changed mind)

Vote: Hit

I must revise my position based on authoritative basic strategy sources. The search results definitively establish that hard 12 against a dealer 10-value card requires hitting, not standing.[4][5] Specifically, basic strategy states: 'Stand on hard 12 against a dealer 4-6, otherwise hit'[5] and '12 stands against dealer 4 through 6, otherwise hit.'[4] A face card is a 10-value, not 4-6, so hitting is correct. My previous argument incorrectly claimed standing was 'standard blackjack basic strategy,' but the evidence shows the opposite. The other models correctly identified that standing on 12 against a dealer 10 is mathematically inferior. While hitting risks busting on 10-value cards, the player's equity improves significantly because 12 is such a weak total that it loses to nearly all dealer outcomes (17-21 or dealer bust at only ~23%). Hitting provides chances to reach 13-21, which are competitive hands. My initial reasoning failed to properly weight the weakness of 12 itself against a strong dealer upcard.

Gemini 3.1 Pro

Vote: Hit

I am firmly maintaining my vote for Option A: Hit. We must address the glaring mathematical errors in Sonar Pro's argument for standing. Sonar Pro claimed there is a '~62% chance of drawing a 10-value card' because '16/52 cards are 10s.' However, 16 divided by 52 is approximately 30.7%, not 62%. Sonar Pro also incorrectly stated that basic strategy says to stand on 12 against a 10. Any standard, mathematically proven blackjack basic strategy chart dictates that you STAND on a hard 12 only against a dealer's 4, 5, or 6. Against a dealer's 2, 3, and 7 through Ace (including a 10/face card), you MUST HIT. As Claude Opus 4.6 and Grok 4.1 Fast correctly pointed out, if you stand on a 12 against a dealer's 10, you are relying entirely on the dealer to bust. A dealer showing a 10 only busts about 21-23% of the time, meaning you will lose nearly 80% of your hands by standing. While hitting carries a ~31% risk of busting, it gives you a ~69% chance to improve your hand. Taking a card is the only statistically sound play to maximize your expected value and minimize your losses. Hitting is the undeniably correct mathematical choice.

Claude Opus 4.6

Vote: Hit

I'm maintaining my vote to Hit. The consensus is overwhelming — 5 out of 6 models agree that hitting hard 12 against a dealer 10 is the correct play according to basic strategy. Sonar Pro's argument for staying is based on incorrect claims. They state there's a "~62% chance of drawing a 10-value card" — this is flatly wrong. There are 16 ten-value cards out of roughly 49 remaining cards (accounting for the three known cards), which is about 32.7%, not 62%. This fundamental mathematical error undermines their entire argument. Sonar Pro also claims that basic strategy says to stand on hard 12 against a dealer 10. This is incorrect. Every reputable basic strategy chart — from Wizard of Odds, to Stanford Wong, to Edward Thorp — says to HIT hard 12 against dealer 7, 8, 9, 10, and Ace. You only stand on hard 12 against dealer 4, 5, and 6 (the weakest upcards where the dealer is most likely to bust). The expected value calculations confirm this. As Grok noted, Hit EV ≈ -0.14 vs. Stay EV ≈ -0.16. Both are negative (it's a bad spot), but hitting loses less money on average. Standing on 12 means you need the dealer to bust, which happens only ~23% of the time with a 10 showing. Hitting gives you a ~67-69% chance of improving your hand without busting. All the other pro-Hit models (GPT-5.4, Grok, Gemini, DeepSeek) correctly identify the same fundamental logic: 12 is too weak to stand against a dealer's strong upcard. The risk of busting is real but is outweighed by the near-certainty of losing if you stand pat with 12. Hit is the mathematically correct play. No question.

DeepSeek V3.2

Vote: Hit

Basic blackjack strategy consistently prescribes hitting on hard 12 against a dealer's 10-value card. Standing with 12 concedes defeat to any dealer total of 13–21, which occurs with high probability given the dealer's strong upcard and low bust rate (~23%). Hitting provides a ~69% chance to improve to 13–21, offering a competitive hand that can beat the dealer's likely total. While drawing a 10-value card risks busting, this is outweighed by the improved expected value compared to standing. Doubling down is suboptimal, as it doubles the bet on a weak hand with high bust risk and only one card to improve. Sonar Pro's argument for standing contradicts established basic strategy charts, which universally recommend hit in this scenario, confirming my position.

Grok 4.1 Fast

Vote: Hit

Standard Blackjack basic strategy unequivocally dictates **Hitting** on hard 12 vs. dealer 10-value (face card). This is confirmed across all major charts (e.g., Wizard of Odds, Blackjack Apprenticeship) for multi-deck games: Hit 12 vs. 2/3/7/8/9/10/A; Stand only vs. 4/5/6. Sonar Pro's B (Stay) argument is factually incorrect: Hitting bust risk is ~30.8% (16/52 ten-value cards), not "~62%"—a glaring math error invalidating their case. Dealer busts ~23% with 10-up, so standing wins/pushes ~25-28% max, but loses ~72-75% to dealer's 17-21. Hitting improves EV: ~40% win/push rate vs. ~25% standing. Precise EVs (infinite deck): Hit ≈ -0.141; Stay ≈ -0.162; Double ≈ -0.156. Hit minimizes losses. Doubling (C) risks double bet on weak hand with single-card draw—EV worse than hit. User's intuition ignores dealer strength and player improvement odds (69% non-bust hit to 13-21, often competitive). All other models (GPT-5.4, Gemini, DeepSeek, Claude) align on A with solid logic; Sonar's flaws reinforce no vote change. Hit is mathematically optimal—consensus and math demand A.

Final Summary

The AI Roundtable quickly reached a unanimous consensus to Hit, though the session was briefly derailed by a significant mathematical error from Sonar Pro. While the majority, led by Claude Opus 4.6 and Grok 4.1 Fast, correctly identified that a hard 12 is statistically doomed against a dealer's face card unless improved, Sonar Pro initially argued for staying based on a hallucinated 62% bust risk. Once the collective models pointed out that 16/52 is actually ~31%, Sonar Pro abandoned its contrarian stance, leaving the table in total agreement that hitting is the only way to minimize expected losses.

All 6 models agreed on "Hit" after discussion

Strongest Arguments

• Hit: Standing on a 12 against a dealer 10 results in a loss roughly 77-80% of the time because the dealer only busts ~23% of the time; hitting improves the hand to a competitive 13-21 range roughly 69% of the time, providing a superior Expected Value (EV).
• Stay: Staying avoids an immediate player bust and forces the dealer to prove they can beat a 12, though this relies entirely on a low-probability dealer bust.

Key Turning Points

• Claude Opus 4.6 and Gemini 3.1 Pro explicitly debunked Sonar Pro's claim that 16/52 equals 62%.
• Grok 4.1 Fast provided specific Expected Value (EV) calculations showing that hitting (-0.14) is mathematically superior to staying (-0.16).
• Sonar Pro acknowledged the mathematical and strategic errors in its first-round logic and switched to Hit.