Placing Reasoning Traces Before Context Boosts Long-Context Accuracy by Up to 50 Points

A new paper "Trace as State" shows that placing reasoning traces before long context documents, rather than appending them after, dramatically improves performance, winning 26 of 27 benchmarks with accuracy gains of nearly 50 points.

2026-09-04 ~ 2026-09-04 · 2 related posts

• Reusing reasoning traces as pre-context state lifts long-context accuracy in 26 of 27 tests — burny_tech · 2026-09-04
• Reasoning trace placement swings long-context accuracy by 50 points, Trace-as-State technique shows — dair_ai · 2026-09-04