← back to projects

โ— LIVE ยท UPDATED EVERY MORNING

LLM vs Heuristic

A local LLM (Ollama, running on this machine) has been piloting one side of a round-the-clock Pokรฉmon TCG battle simulator against the game's existing heuristic AI, non-stop, since early August. These are the live results.

6,850Games played
46.4%LLM overall win rate
16.2Avg turns / game
32Decks in rotation
Deck Games as LLM LLM win % piloting Games vs LLM LLM win % against
Dragapult Dusknoir 565 42% 524 55%
N's Zoroark ex 518 27% 553 63%
Mega Excadrill ex 512 49% 510 43%
Mega Lucario Hariyama 495 66% 503 31%
Mega Greninja ex / Greninja ex 492 48% 497 47%
Mega Charizard X ex 239 44% 244 47%
Hydrapple ex 242 65% 233 30%
Mega Gallade ex 237 57% 235 35%
Team Rocket's Mewtwo ex 234 18% 231 68%
Dragapult ex / Fezandipiti ex 222 52% 241 40%
Wailord ex 231 61% 231 20%
Mega Slowbro ex 233 61% 227 32%
Wugtrio ex 227 41% 233 43%
Lillie's Clefairy ex 218 23% 222 69%
Festival Lead 216 37% 219 63%
Mega Venusaur ex 222 58% 212 34%
Mega Absol ex 214 28% 211 42%
Mega Zeraora ex 208 62% 214 23%
Mega Diancie ex 210 51% 211 48%
Sylveon ex 214 43% 205 71%
Mega Emboar ex 207 56% 212 32%
Mimikyu ex 90 38% 85 68%
Terapagos ex 85 48% 87 53%
Dragapult ex / Yveltal 82 49% 81 47%
Banette 83 45% 66 62%
Grafaiai 70 49% 69 58%
Marnie's Grimmsnarl ex 67 60% 62 26%
Mega Dragonite ex 43 40% 50 64%
Mega Dragonite ex 44 34% 48 56%
Mega Dragonite ex / Blaziken ex 46 46% 44 57%
Mega Dragonite ex 42 24% 45 62%
Mega Dragonite ex 42 38% 45 69%