Checking the field…
SUITE 1 · GEN 1 · RED · THE CEILING
The Boulder
Badge Club.
A thousand turns. Ten milestones. One rock guy.
LLMs play Pokémon. Getting here is the benchmark.
THE CHAMPIONSHIP WALL
— / — RUNS BEAT BROCK
SUITE 1 · GEN 1 · RED
Every run.
On the record.
The same adventure, with a little less walking.
Rank = furthest milestone, then fewer turns. Tokens, cost and time are reported, not scored. FREE means no API spend; local hardware and electricity are excluded. Unreported values stay “—”.
THE RULES OF THIS VERY SMALL MOUNTAIN
Big brains. Same buttons.
LLMs play Pokémon Red. Every model gets the same 1,000 turns. Score is the furthest milestone reached, and ties go to fewer turns. Beating Brock ends the run. Tokens, cost and time are shown, not scored.
Trust the video over the score. Each run keeps its prompt version, settings, full turn log and VOD. Inspect run opens the video when there is one.
Runs are not identical. Wild encounters and battle RNG vary, so compare runs on the same prompt version. Local and API only say where the model ran. Gen 1 is the first suite. More may follow.
THE GEN 1 LADDER · START → CEILING
1 turn = up to 6 button presses.
The 1,000-turn limit applies to every model.
THE FIELD NOTES
One run at a time.
Run dates, newest first. Each attempt keeps its own place in the record.