How GTO poker training works here

Getting started

GTO Academy is a GTO poker trainer you play rather than watch. It deals you the spot, you act, and a solver prices what you did in big blinds. Everything else on this page is detail on top of that one sentence.

Three things to do first, in order:

1. Pick the game you actually play. Tournaments, cash or sit & go — each is its own path, because a table size and a depth ladder are what a path pins, and finishing the tournament syllabus does not teach the 6-max cash game.
2. Play chapter 1. Forty graded decisions at 78% clears it. It is the real chapter, graded against the same solved reference a subscriber trains on, and it costs a free account and no card.
3. Bring a session back. The loop only closes once the trainer has seen hands you played for real — that is what turns the last three chapters from a syllabus into your own homework.

The loop, end to end

Most GTO poker tools are an instrument: a solver, a chart, a report. An instrument leaves you holding the question you arrived with — what should I do with the next ten minutes. This one is a circuit, and the circuit is the product:

Then round again — and the fourth step is what makes it a loop rather than a list. 3 of the 17 chapters have no content of their own until they have read your own graded history.

01 — Learn: the trainer deals the spot

The trainer is a syllabus, not a form. It opens on a path — 17 chapters in a fixed order, 14 of them taught and 3 built from your own history — and each chapter pins the seats, the stack depths and how much of the hand you play. You are dealt a real spot from that chapter, you act, and the answer comes back priced.

Three paths, because three games

A track fixes the table size and the depth ladder, so the same chapter number means a different table on each one:

• Tournaments. Full-ring MTT, 8-handed, from 40bb down to a shove stack. Open, defend and shove correctly at every depth a tournament puts you through — and know which of your seats is costing you the most.
• Cash game. 6-max, deep. Every seat at 100bb, then the 50bb short-stack game. A complete 6-max game: opening ranges, blind defence, and postflop lines priced in big blinds rather than guessed at — deep, and again at half the depth.
• Sit & Go. 3-handed hyper-turbo. Short, sharp, and mostly preflop. The three-handed endgame: which hands open, which defend, and where the shove threshold really sits at 25, 15 and 10bb.

What a chapter asks of you

Chapters are counted in spots — graded decisions — rather than hands, because a preflop-only hand asks one question and a hand played to the river asks five, and calling both "a hand" would make the same goal mean two different amounts of work. A chapter names its count and the accuracy it wants, and clearing it opens the next one.

Prefer to set the table yourself? Build your own spot takes a format, a table size, a stack depth, your seat and the action in front of you, and deals from exactly that. It is the same grading, without the syllabus choosing for you.

02 — Play: at your own poker room

Nothing runs beside your tables. There is no HUD, no overlay and no client to install — which is also what keeps this the right side of every poker room's terms. Your room already writes a hand history, and that file is the whole of what step 3 reads.

Rooms arrive in one of two ways, and which one yours is decides your chore:

Saved to a folder as you play

The client writes every hand into a folder on your machine as you play, once the setting is on. Connect that folder once and it is re-read after each session.

• PokerStars — Saved as you play, once you switch it on
• ACR — Saved as you play, once you switch it on
• Winamax — Saved automatically, with nothing to switch on
• 888poker — Saved as you play, once you switch it on

Downloaded from the room

The room keeps the hands on its side and hands them over as a file you fetch, sometimes a day later. Those get dragged in.

• GGPoker — Downloaded from PokerCraft as a .zip
• partypoker — Exported from My Game, a day after you play
• Ignition — Downloaded from the client, 24 hours after you play
• CoinPoker — Requested from CoinPoker support

The exact menu path and folder for each of these 8 rooms is on the import screen, next to the drop zone. Room logos belong to their owners; GTO Academy is not affiliated with any of them.

03 — Analyse: drop the session in

Drop the file in and each sitting lands with its result, its length and its hands. A session is a row you can open; a hand is a replay you can step through.

What the reader will and will not claim

Reading a hand history is where a tool could quietly lie to you, so the rules are strict and the screen states them:

• A hand fails, never the file. Every hand that cannot be read is counted under a named reason. "412 of 415 imported" with no account of the three is the kind of quiet wrong answer you find out about three weeks later.
• Hands you already have are counted, not dropped. Re-reading a folder is the normal case — a client appends to one file all session — so on most imports that is the biggest number on the screen.
• A snapped bet size says it snapped. A 1.7bb bet graded on the tree's 1.8bb branch is reported as such, because the EV answers a slightly different bet from the one you made.
• An unmatched line stops the walk. If the solve tree has no branch of that kind, the review stops with a reason rather than snapping to the nearest thing. Streets after it are not shown, because seeding them would mean inventing a range the hand never had.
• A hand with no row is listed and not scored. Printing "✔ 0.00bb" for a combo the range never held would be the product's most confident lie.

And an AI coach reads the same report

The solver gives you numbers; the coach turns one spot's numbers into a sentence. It is grounded in the solved node itself — every action's frequency, your combo's frequency and EV, the pot, the stacks, the equity, the solve's own error bar — and it is told it may not state a number that is not in that report. It never runs on arrival: it is a button, and the button says what it spends before you press it.

Where your hands live

In your browser. Files are read locally, hands are stored on your machine, and the solve runs on your own CPU — nothing is uploaded. The trade, stated plainly: the hand database is per browser, so importing on a laptop does not put the hands on your phone. The leaks learned from them do follow you, because those are rows on the server.

04 — Improve: drill what it found

A course is the same for everybody. Remedial work never is. 3 chapters of every path — Your preflop leaks, Your postflop leaks, Your leaks, again — have no content of their own. When you reach one, it reads your own graded history for that track and pins the spots you are actually worst at as a short list of drills. Clearing the list is what finishes the chapter.

Leaks are ranked by what they cost rather than by how wrong they look: a spot you get right half the time over fifty decisions outranks one you get right a fifth of the time over eight. That is the whole reason the loop is worth closing — the fourth step cannot exist until the third one has read real hands.

The same history is what the analysis screen reports on, and it counts decisions from both sources: hands the trainer dealt you and hands you played for money. They are the same kind of evidence.

How a decision is graded

A GTO poker strategy mixes. In one spot the solver 3-bets a hand 60% of the time and calls it 40%, and both are the strategy — so "correct" cannot mean "the single best action". Two rules decide every verdict in this product:

1. Frequency decides the tier. What the solved strategy does with your exact combo in that exact node is what grades you. An action the solver skips is at best an inaccuracy, however small the loss.
2. EV only prices it. The big-blind figure tells you what the choice cost. It can demote a verdict; it can never promote one.

The second rule matters more than it looks. At equilibrium every action inside a mix has the same EV by construction — equilibrium is indifference — and an action just outside the mix is only slightly worse. So an EV gap cannot tell "you found the mix" from "the solver never does this", and letting one overrule the frequency graded lines the solve takes 0% of the time as correct.

• Thresholds are fractions of the pot, not absolute big blinds: 0.05bb is 0.9% of a small single-raised pot and almost nothing in a 3-bet pot, so one constant would be making two different claims.
• Nothing is graded below the solve's own error bar. Every threshold is floored at the solve's reported exploitability — a gap smaller than that is the solver's noise, not your mistake.
• The headline and the number agree. When a frequency rounds to zero the verdict names what the chart actually does instead of quoting a zero at you.

The solver

The same engine that grades you is open as a tool of its own at /solver. Describe the spot — the preflop action, a board, the pot and the effective stack — and it solves it in front of you, then prices every line for every hand in the range.

It runs client-side, in a Web Worker on your own CPU. There is no download, no account and nothing to upload, which is also why the free allowance here is a courtesy rather than a lock: there is no compute of ours to withhold. 5 guided walkthroughs are free, and a subscription lifts the count.

Preflop charts

Most hands in a real session never see a flop, so the preflop charts are the part of this product a player uses most and pays for least. Every hold'em chart at /charts is free and needs no account: opening ranges by seat and depth, defending and 3-bet ranges, and push/fold charts for short stacks.

A cell that is part green and part blue is the thing a static chart image cannot show and the thing beginners get wrong most: the solved answer to that hand is both, in those proportions. Pot-Limit Omaha has its own charts at /plo/charts — the spot builder is open to everyone, the solved answers are part of the subscription.

What is free and what Pro adds

No card is involved in the free tier at any point, and there is no hand ceiling on it — chapters 1 and 2 are unmetered. The figures, the currencies and what a subscription renews at are on the pricing page; cancelling needs no sign-in and takes effect at the end of the period you have paid for.

Where your data lives

• Hand histories never leave your machine. Files are read in the browser and stored in it. A grinder's thirty thousand hands a month have no reason to travel, and the solve that prices them runs on your own processor.
• What does reach us is one row per graded decision — the spot, what you did, how it graded. That is what the leak report and the adaptive chapters are built from, and it is why your progress follows you between devices while the hands do not.
• The AI coach is the one thing fetched rather than computed. It is sent a validated report of one solved node, never a file and never your history.
• Folder access is yours to grant and to revoke. Chromium browsers can re-read a connected folder each visit; Safari and Firefox have no such permission, so those files get dragged in. Nothing imports while the tab is shut.

The full text is in the privacy policy.

The GTO vocabulary this product uses

GTO in poker is short for game theory optimal: the strategy that cannot be beaten in the long run by any counter-strategy, because it defends every part of its own range. Every screen here assumes five words, and this is all five:

• Range. Not a hand — the whole set of hands a seat can hold here, with how often it holds each. The 13×13 grid is a range drawn out.
• Mix. One hand played more than one way, in fixed proportions. A mixed cell is a solved answer, not an undecided one.
• EV, in big blinds. What a line is worth on average, measured against the alternatives. Every verdict here is priced this way because big blinds compare across stakes and dollars do not.
• Exploitability. How far a solve still is from equilibrium, in big blinds. It is the solver's own error bar, and nothing is graded below it.
• Exploitative play. Deliberately leaving the GTO strategy to punish a specific opponent's mistake. Two chapters of every path are about exactly this — GTO poker is the baseline you deviate from, not a script you follow forever.

Longer versions: what GTO in poker means for the theory, and the poker glossary for every other term, with the probabilities.

Frequently asked questions

Start the loop

Reading about a GTO strategy and being able to play one are different skills, and only one of them is built by reading. Chapter 1 is forty graded decisions, it needs a free account and no card, and it is the shortest route from this page into the loop it describes.