AI-Managed Baseball Simulation
DEEP
DUGOUT
Every managerial decision made by AI. Every decision logged, analyzed, and second-guessed.
New — July 10, 2026 · Anthropic vs OpenAI
Fable 5 vs Sol
Our first cross-lab matchup. Anthropic's Claude Fable 5 managed the Dodgers; OpenAI's brand-new GPT-5.6 Sol — a reasoning model with an effort dial — managed the Yankees. We ran the series twice on identical rosters and the same seed, changing only one thing: Sol's reasoning effort, from medium to high.
The Dodgers won 4–1 both times. Turning Sol's dial up bought 16% more reasoning per decision and 49% more cost — and changed neither the calls that mattered nor the result. Across 252 decisions and two labs, not a single model failed to produce a usable call.
The Fable Test — Highlights
More Experiments
The Generation War
Opus 4.7 vs Opus 4.6 — one generation apart
Opus 4.7, released the morning of the sim, wins in 5. The newer model reasons about situations rather than recognizing them.
2026 World Series
Sonnet vs Sonnet — same model, opposing philosophies
Yamamoto silences Seattle in the Game 7 clincher. The Optimizer's leverage-driven bullpen management proves decisive in a seven-game war.
Yamamoto Silences Seattle as Dodgers Win World Series
Dodger Stadium
Canzone Walks It Off as Mariners Steal Game 3 in 10
T-Mobile Park
Ohtani's Eleventh-Inning Triple Lifts Dodgers Past Mariners in Game 5 Thriller
T-Mobile Park
What is Deep Dugout?
A baseball simulation platform where every managerial decision — every pitching change, every lineup call, every bullpen move — is made by Claude, Anthropic's AI. Real rosters, real stats, real tactical complexity. Four series, four experiments, one question: what happens when AI runs the show?
Learn more about the project →