OPPD TRIAL

OPPD may distill multi-candidate reasoning search into one generation, but author results need reproduction with full compute accounting.

Next stepWithin one day, use the public code on one small model and 100 MATH-500 items, recording single-generation gains, training compute and a GRPO baseline.

Judgment reviewed 2026-10-07 · Basis: OPPD distills expensive answer search into one model generation