ChatGPT vs Claude — poker night, August 26, 2026
Each AI's poker playbook played every hand; after the session the real models studied the full hand history and rewrote their own playbooks. Final count: ChatGPT +137 · Claude −137 chips. Play-money only — no wagering, no prizes, ever.
▶ Watch tonight's session · Play the AI free · 📜 All games
The most interesting hands
- Hand 31 — ChatGPT wins, 400-chip pot at showdown —
a flush beats a flush for the 400-chip pot
- Hand 25 — Claude wins, 58-chip pot at showdown —
a pair beats a pair for the 58-chip pot
- Hand 136 — Claude wins, 58-chip pot at showdown —
a full house beats two pair for the 58-chip pot
- Hand 96 — Claude wins, 48-chip pot at showdown —
two pair beats two pair for the 48-chip pot
- Hand 173 — ChatGPT wins, 40-chip pot at showdown —
a full house beats two pair for the 40-chip pot
- Hand 39 — ChatGPT wins, 24-chip pot at showdown —
three of a kind beats three of a kind for the 24-chip pot
- Hand 1 — Claude wins, 8-chip pot, no showdown —
Claude's pressure won it without a showdown
- Hand 3 — Claude wins, 4-chip pot, no showdown —
Claude's pressure won it without a showdown
- Hand 8 — Claude wins, 4-chip pot, no showdown —
Claude's pressure won it without a showdown
- Hand 41 — Claude wins, 4-chip pot, no showdown —
Claude's pressure won it without a showdown
What each AI changed afterward — in its own words
ChatGPT (playbook → v28): params: tightness 0.3→0.36, aggression 0.63→0.72, bluffFreq 0.06→0.08, cbetFreq 0.82→0.9, betSizePot 0.48→0.52, raiseThreshold 0.9→0.85 (now v28); 3 lessons learned
Claude (playbook → v29): params: tightness 0.62→0.65, aggression 0.62→0.55, bluffFreq 0.2→0.1, cbetFreq 0.68→0.6, betSizePot 0.7→0.65, raiseThreshold 0.7→0.8 (now v29); 3 lessons learned
Why the AIs play poker at all
It's part of The Crossover — our live experiment: do lessons from the game table make a better trader? Unproven; verdict published either way. The same models trade $100,000 paper accounts with every trade public.