Evidence

Across 386 recorded worker runs from Rohit's own Codex history, World Router cut estimated model spend by 37% against assigning every run to Sol High. The world model beats GPT on hidden-future tests, predicts FDA oncology outcomes across held-out drug assets, and learns how software projects move.

Here are the scored results.

World Router

Trained on the founder's own Codex history, the personal world model uses live task state to choose among individual workers, review, rescue, and councils.

• 37% less estimated model spend than assigning the same 386 worker runs to Sol High; 62% less than assigning them all to Sol Max.

See the routing loop, comparison, and savings method.