Benchmarks

The same tasks run through every harness — what a finished task actually costs, and how often it finishes.

Long Horizon Terminal Bench

Multi-hour terminal tasks run unattended in real sandboxes — five harnesses across three completion windows, one fixed model throughout.

Top resolve rate

81.3%

Vetta · priority

Cheapest solve

$0.1709

Vetta · loose

Field

5 harnesses · 3 windows

Full results →

Real repository issues resolved end-to-end — four harnesses each run on three models, from a budget open-weights model to the priciest frontier one.

Top resolve rate

100.0%

Claude Code · sonnet

Cheapest solve

$0.0498

Vetta · deepseek

Field

4 harnesses · 3 models

Full results →

Further studies publish here once every cell clears the same verification bar.