Overpair Bet Sizing: When the Big Bet Costs More

You raise the button with A♠ A♥, the big blind calls, and the flop comes 7♠ 5♠ 3♥. The big blind checks. With straight possibilities and two spades on the board, a substantial bet can feel like the obvious way to protect your aces.

In this GTO Gecko library export, betting 4.1bb into 5.5bb gives up 0.282bb compared with checking. Betting 1.8bb has a stored gap of 0.003bb. The aces check 90.9%. A small bet is still close in stored value; the larger bet is not.

On a low two-tone flop, compare a high overpair’s small bet with its check before reaching for the larger size. That comparison matters most when the hand checks heavily. Keep reading for a board where queens should usually bet instead.

These are stored solver outputs retrieved September 11, 2026. Solver version and achieved convergence are unavailable, so small EV gaps have no verified accuracy bound. Some actions retain small positive frequencies despite nonzero stored loss, so these are approximate outputs rather than exact equilibrium mixing conditions. The figures show observations and calculations; our explanations of the ranges are interpretations. The header is AI-generated editorial artwork.

The exact decision

The study uses six-max no-limit Hold’em cash models, 100bb before posting, no ante, and 5% rake capped at 4bb. The button opens to 2.5bb and the big blind calls. The folded small blind contributes the remaining 0.5bb to the 5.5bb flop pot.

The big blind has only a check at the exported root. The button can check, bet 1.8bb or bet 4.1bb—about 33% or 75% of the pot. “Large” below means the larger of those two offered sizes. Other sizes and a flop lead were not tested.

The first row is the opening example: 6.087 − 5.805 = 0.282bb for the larger bet, versus 6.087 − 6.084 = 0.003bb for the smaller one. A frequent check does not make every bet a substantial mistake.

Compare the eights with the aces. The lower overpair bets much more often, yet several alternatives remain close in stored EV. Neither the frequency ladder nor the presence of a spade supplies an automatic action for every overpair.

How often did the large-bet shortcut fail?

We explored eight flop classes, then froze the test before opening twelve more from six new rank patterns. The button sample contains 564 exact overpair decisions across 20 boards. Ten boards are two-tone and ten rainbow; these are chosen classes, not a random sample of flops dealt in play.

The shortcut under test was specific: bet 4.1bb with every overpair on a two-tone flop. We called it locally cheap on a board only if at least 95% of its overpair range lost no more than 0.10bb against the stored continuations. We measured each action against the highest available EV at that same decision. The 0.10bb threshold is a study screen, about 1.8% of this pot, not a universal error definition.

Large-bet loss above 0.10bb
Share of overpair range on each two-tone flop. Dashed marker: the 5% failure line.

9♠ 8♠ 7♥ 3.3%

Discovery · 30 exact overpair combos

T♠ 9♠ 8♥ 0.0%

Discovery · 24 exact overpair combos

8♠ 7♠ 6♥ 8.3%

Discovery · 36 exact overpair combos

7♠ 5♠ 3♥ 35.7%

Discovery · 42 exact overpair combos

K♠ 8♠ 3♥ 0.0%

Held out · 6 exact overpair combos

Q♠ 9♠ 4♥ 0.0%

Held out · 12 exact overpair combos

J♠ 7♠ 2♥ 0.0%

Held out · 18 exact overpair combos

T♠ 6♠ 2♥ 0.0%

Held out · 24 exact overpair combos

6♠ 5♠ 4♥ 18.8%

Held out · 48 exact overpair combos

5♠ 4♠ 3♥ 5.6%

Held out · 54 exact overpair combos

The failed boards are 7♠ 5♠ 3♥, 8♠ 7♠ 6♥, 6♠ 5♠ 4♥ and 5♠ 4♠ 3♥. The shortcut fails this per-board screen. The last board is close to the boundary: 5.6% exceeds the loss threshold, against a 5% cutoff. The exact failure count depends on that choice; the sensitivity results retain other loss thresholds. The average loss is modest: 0.022bb across the ten two-tone board classes, with each board weighted equally. The largest observed large-bet loss is the opening aces example at 0.282bb.

As a secondary, exploratory finding, the smaller bet stayed within 0.061bb of the highest stored action EV in every one of the 564 button overpair rows. This is the practical reason to compare sizes: frequent checking often leaves a cheap betting alternative available. It does not establish the cost of a whole strategy that always bets small, or prove the small gaps would survive a more accurate solve.

Read the range that responds

“Charge the draws” names one possible benefit of betting. It leaves out which hands fold, which call, and which raise. GTO Wizard’s protection analysis makes that broader comparison in a different, 200bb blind-versus-blind setup. The numbers here come from our 100bb button study.

On 7♠ 5♠ 3♥, the big blind’s range contains four weighted combinations of 64s, plus sets and two-pair hands. The button has more overpair mass, but that alone cannot describe how its aces perform after a particular bet.

The first grid shows BTN’s complete flop decision; the second shows BB’s response to the 4.1bb bet. Entry weights include the preflop ranges and flop card removal. BB’s forced check removes no additional hands. Both grids are visible before any interaction; use the hand selector or arrow keys for exact class values.

We keep BB’s stored response to the button’s betting range and remove combos containing either of the specified aces. This changes the weights used in the table; it neither gives BB knowledge of those aces nor re-solves a response against them.

With the aces fixed, the bigger bet gets more folds: 52.2% versus 31.9%. It also gets fewer raises, 12.7% versus 17.4%, though the raise is to 11bb instead of 6.4bb. More folds and fewer raises still do not make the larger bet better for this hand.

Most of the large bet’s folding mass—94.4%—is currently unpaired. Direct flush or straight draws account for 34.5% of the smaller bet’s calling mass and 44.7% of the larger bet’s. Those shares describe different calling ranges, not an increase in the number of draws calling. Draws can overlap pairs; already-made straights count as draws only when they also have a flush draw. This describes the filtering without identifying each branch’s contribution to the aces’ EV.

Two comparisons that stop the rule going too far

A flush draw is not required for the large bet to lose value. With A♠ A♥ on rainbow 6♠ 5♥ 4♦, betting 4.1bb trails checking by 0.134bb. On two-tone 6♠ 5♠ 4♥, it trails by 0.138bb. That is a similar result on two different suit textures.

Now switch to queens on J♠ 7♠ 2♥. Both bet sizes beat checking by about 0.19bb. The rainbow version rewards betting too. “Check overpairs on wet boards” would replace one faulty shortcut with another.

Four additional cutoff-versus-big-blind trees use different entering ranges. On 7♠ 5♠ 3♥, the cutoff’s A♠ A♥ also prefers checking to 4.1bb by a material stored gap, 0.263bb. We keep those 126 overpair rows separate: changing opener position changes the ranges, so this is a comparison, not an isolated position experiment.

Nor does the result transfer automatically to a three-bet pot, another stack depth or a tournament. Upswing’s overpair discussion gives a useful opposing case: the usual higher-pairs-check-more pattern can reverse when the in-position player called a three-bet and faces a strong checked range.

A three-hand study exercise

Cover the EV columns or write down your answers before reading the tables. Predict the action for aces on 7♠ 5♠ 3♥, aces on 6♠ 5♠ 4♥, and queens on J♠ 7♠ 2♥. For each, predict which option loses the most value and which alternatives are close. Compare check with both bet sizes, then reveal the EVs and inspect the frequencies. Explaining the largest loss is more useful here than guessing an exact mixing percentage.

Keep this conditional cue: when a high overpair checks heavily on a low, connected or two-tone flop in this model, retain the small bet in your comparison. Check the actual range response and exact hand. A draw-heavy board alone is too little information to select the large bet, and a frequent check is too little information to reject the small one.

Our slow-playing casebook separates the chance of being outdrawn from the cost of checking. This study adds the missing action-value comparison for a particular flop model. GTO Gecko’s paid study plans let you compare precomputed actions, frequencies, EV estimates and ranges in available scenarios; the current app listing describes access by plan.

Method and limits

All 24 requested study trees were retrieved: 20 BTN boards and four CO comparisons. An overpair is a positive-weight pocket pair above the highest flop rank. Local action loss is max(available action EVs) − chosen action EV. Within a board, we weight exact combos by exported marginal reach; across a named board group, each class receives equal weight. These are not deal frequencies, confidence intervals or bb/100 estimates.

Probabilities and reach are rounded to two decimals. Rows totaling 0.99 or 1.01 were normalized for frequency calculations. BB response summaries additionally remove combos conflicting with the specified hero cards. EVs were converted to bb using the current app’s scale; three displayed decimals reflect export resolution, not proven accuracy.

The selected sample, fixed rake and stack, imposed BB check, limited bet menu and unknown convergence restrict the conclusions. We did not re-solve, remove raises, evaluate a changed full strategy, measure showdown equity, or model opponent mistakes. Our solver-convergence exercise explains why numerical output needs an accuracy context.

Download the selected exact-hand values, 24-tree board results, response composition and method, thresholds and arithmetic. Complete proprietary trees remain private. For the distinction between range weights and matchup frequency, see Pio’s number definitions.