Cycle 00000000-0000-4000-8000-000000002609 · Index v2.0 · published

Demo data · not real results

What changed in September 2026 (demo)

Top-rank quality at 1/4 the cost per successful task. 5 models share rank 1 (none is significantly better than another on the same tasks); Model F · low effort (Index 93.6) costs $0.028 per successful task, Model D · high effort (Index 94.5) $0.13.

28 Sep 2026data through
2 Oct 2026published
1,560runs
13entries
1new entries
1changed rank

01Movement

Against August 2026 (demo)

Climbed

  • Model H▲1 to 9–13 · +1.4 Index

Slipped

No entry lost rank.

New this cycle

  • Model JEnters at 9–13 · Index 46.5

02The board

Index against cost per successful task

Navigator Index (0–100) — higher is better
Cost per successful task (log scale) — lower is better
Navigator Bench · Index v2.0 · data through 28 Sep 2026Line: Pareto frontier · quadrants split at Index 70 and the median $0.014Demo data

September 2026 (demo): Model K · low effort was the cheapest model at Index 70 or better, at $0.0056 per successful task. August 2026 (demo): $0.0057.

← August 2026 (demo)October 2026 (demo) →Full board for September 2026 (demo)ChangelogMethodology

Navigator Bench · Index v2.0 · data through 28 Sep 2026