Leaders
The best open-weight models, and the closed ones to measure them against
The first three cards are closed models: the highest score, the lowest cost per case, and the strongest proprietary model from China. The strongest open-weight models follow, one card each, best run first. Every run is in the chart and the table below.
Chart
Score and cost, every model
Runs added in the October 2026 release are marked New. A model can appear more than once: the tool path column says which run is which. To compare agent harnesses, a trophy marks the best proprietary run in each of Louie.ai, Claude Code, Codex and opencode.
Run-specific caveats
Table
All open-weight runs
Click a column heading to re-sort.