PokerToast · study tools for mixed-game poker
About PokerToast and its solvers
PokerToast is a study tool for mixed-game poker (Razz, Seven Card Stud, Stud 8, limit Hold'em and Omaha 8): its own trained solvers, a hand and range calculator, practice tables against bots with a solver review of your decisions (every spot the solver has learned; it tells you when it cannot judge one), and a daily solver-graded spot. There is no real-money play.
Who makes it
PokerToast is a small, independent study site run by one mixed-game player in the United States. The owner's legal details are in the Terms and Privacy Policy. Questions and suggestions are answered by the owner himself: contact@pokertoast.com.
How the solvers were trained
Each game has its own solver, written for PokerToast and trained by self-play with Monte Carlo counterfactual regret minimization (external-sampling MCCFR), the standard method behind poker solvers: the program plays itself millions of times and, in every situation it meets, keeps shifting toward the plays that would have done better. The result is a strategy that says how often to fold, check, call, bet or raise in each situation.
- Betting: limit, at most four bets a street. Stud games use 20/40-style stakes (ante 0.15 small bets, bring-in 0.25); Hold'em and Omaha 8 use blinds of 0.5 and 1 small bet. One simplification in the stud games: on 3rd street, when the other player just calls the bring-in, the street ends; under real rules the bring-in could still complete. So spots where a call is close to completing lean a little towards calling. In Seven Card Stud the bet on 4th street is always the small bet, also with an open pair (the standard rules of many live games allow the big bet there).
- Pot sizes: each solver groups the pot before the street into three bands (small, medium, large), so its numbers for a spot are averages over that band; in the largest band the values can be bigger than one particular pot could pay.
- Players: heads-up (two players) for every game; 3-handed solvers exist for all five and are quoted where they are loaded (the table below); at 3 or more players, Stud, Stud 8 and Omaha 8 grades and hints are off for now (those 3-handed solvers are not accurate enough yet), so the bots play them but your decisions are not judged. There is no solver for four or more players.
- Hand grouping: a solver cannot hold every exact deal, so similar hands are grouped (for example in Razz: a made 8-or-better low, or how many cards you have toward one and how many of the cards you need already show on the opponent's board; in Hold'em: the 169 starting hands, then how your hand uses the board and the board's texture). Each game's page says what its grouping looks at.
- EV ("worth"): after training, each play's value is measured by playing the situation out many times with the finished strategy: small bets won or lost from that decision on, compared with folding (0). Pages show how many play-outs a number rests on where it matters.
- Checks: automated tests compare each solver's hand rules with PokerToast's own hand evaluator on thousands of showdowns per game, and the JavaScript copy of each grouping with the Python original on thousands of cases, so the advice the site shows is the strategy that was trained. A training run ends with test matches against simple bots.
- An independent check: Cepheus, the University of Alberta program that essentially solved heads-up limit Hold'em (Bowling et al., Science, 2015), published how often its button folds before the flop; PokerToast's own number stands next to it on the Hold'em Reference pages, so you can compare the two.
What is loaded on this server right now
| Game | Heads-up solver | 3-handed solver |
|---|---|---|
| Razz | in use (35,165 situations, 3,690,500 iterations, trained 2026-10-05) | in use (172,805 situations, 25,358,751 iterations, trained 2026-10-05) |
| Seven Card Stud | in use (30,546 situations, 9,747,144 iterations, trained 2026-10-06) | in use (150,064 situations, 14,496,030 iterations, trained 2026-10-05) |
| Stud 8 (Stud Hi-Lo, eight or better) | in use (47,383 situations, 7,862,000 iterations, trained 2026-10-05) | in use (301,198 situations, 17,428,140 iterations, trained 2026-10-06) |
| Limit Hold'em | in use (63,797 situations, 9,047,710 iterations, trained 2026-10-07) | in use (158,796 situations, 10,703,200 iterations, trained 2026-10-06) |
| Omaha 8 (Omaha Hi-Lo, eight or better) | in use (93,573 situations, 19,095,762 iterations, trained 2026-10-05) | in use (582,245 situations, 16,678,240 iterations, trained 2026-10-06) |
"Situations" is how many distinct grouped situations the strategy holds; "iterations" is how many self-play iterations trained it. A retrained solver changes these numbers and every page that quotes it.
What the numbers can and cannot tell you
- Every solver number is an estimate. It is good for the spots the solver met often, and noisier where hands are grouped coarsely or the spot is rare.
- The numbers are for the hand group, not your exact cards.
- The numbers are for heads-up play unless a page says 3-handed. Most published advice is for full-ring games, so where PokerToast and a book disagree, part of the gap can be the number of players.
- "Worth" assumes an opponent who plays like the solver, not like the players at your table.
- Use the numbers to study. Do not use them during a real-money game: most poker sites ban real-time assistance, and PokerToast is not built for it.
Where to see it
- The Mixed-Game Reference: common questions answered with the solver's numbers, the method and the common advice.
- One page per game: Razz, Stud, Stud 8, Hold'em, Omaha 8.
- Free equity calculators, worked out on the server, no sign-up.
- The Daily Spot: one solver-graded decision a day per game.
- PokerToast itself: the Research Calculator and the Bot Lab (free account).
- Help and FAQ, Privacy Policy, Terms of Service.