we rented 3 GPUs to run 16 AI agents at once so you don't have to. the H100 lost memory, not compute, decided it

we've been testing this in championship manager 97/98: every AI manager handed a random formation, 15 worlds, original 1997 squads. top three are level, 4-5-1 defensive costs ~9 pts a season, and no formation counters another. what does change it is your own strength

grok made 3,262 tool calls and thought for 14.8 h. sonnet made 218 in 36 min the two local models got the most calls rejected: gemma 18%, qwen 11%. the cloud ones stayed under 4%

final scorecard (league pts while in charge + 5 per season survived): opus 92 · grok 86 · deepseek 24 · gemma 23 · sonnet 23 · qwen 21 one run each so there's luck in it, next runs repeat every model

season 1: grok 3rd (77 pts), opus 4th (74). the game's own AI with the same squad got 78 and 73 so the best models matched a 1997 game AI. none beat it

it's over. grok got sacked on 15 sep 1998, 16th with 4 pts from 5 games. all 6 AIs have now been fired from atlético how long each one lasted 👇

grok 4.7 managing atlético in cm 97/98. copa semi, 2nd leg v real madrid, 0-0 from the first leg. every tool call it made, from the team sheet to the final whistle. #Ai #AIagents #retrogaming

what went wrong in this run, all being fixed for the next one: 1. the "available for transfer" flag the tools show doesn't match the game's own filter. 22 of Opus's bids and 19 of Grok's came back "not for sale at any price"

how the CM 97/98 AI league actually works, since a few people asked 🧵 the game is the original DOS version, unmodified, in DOSBox-X. 6 leagues run at once, each one its own emulator on a virtual display, all from the same July 1997 save

been running grok 4.7 as a football manager in a 1997 game for 2 days, only model the board hasn't sacked yet