Cash, SNG, MTT — how the same bots score in different formats

I'd give you a single tier list and tell you that's the answer, but it'd be a lie. Three bots out of the 12 I tracked are dual-format — they play cash and SNG with the same model — and they score differently in each. That's the whole point of this page.

Cross-format scoring table

Bots labelled Bot A through Bot L to keep this clean. The labels are stable across this site so you can follow them.

What jumps out

Cash scores are higher on average. Cash mean across the bots I tracked sits at 5.9. SNG mean is 6.0, but the floor is higher (no sub-4 SNG bot in my sample), so SNG looks more uniform. MTT mean is 4.9 — that's the format with both the lowest ceiling and the most variance.

Dual-format bots are not equally good in both. Bot B is the clearest example. As a cash HU bot it's a 7. As a HU SNG bot it's an 8. Why? Because its end-game push-fold logic is rock-solid — and that's 80% of what SNG is. In cash, the same engine has to make harder turn and river decisions, and it overfolds turns by about 6 percentage points relative to where the solver wants it. Same code, different exposure to its weaknesses.

Three bots flat-out refuse to scale to MTT. Their ICM layer is either missing or broken. I watched Bot D get-it-in 32bb effective with a hand the ICM model said was a clear fold. Twice in one session, identical spot. Not a coincidence, a hardcoded chip-EV decision masquerading as tournament play.

Which format would I personally play a bot in?

If forced to pick one battleground I'd play HU cash. The good cash bots are scary, but the bad ones are really bad and easy to spot, so EV variance across opponents is high. SNG bots are tighter as a cluster — harder to find a weak one, harder to crush the strong ones. MTT HU is the most exploitable on average but the rake/structure usually eats the edge unless the buy-in is meaningful.

What I'd score next

Two of the bots that didn't make this list (one cash-only, one SNG-only) are getting sample-size catch-up runs through July. If anything moves I'll annotate this page — I don't rewrite history, I add a dated note. See how I score for the methodology and why I won't publish a score before 600 hands.