local venue · consumer GPUs

What can your own GPU actually run?

The hardware side of the benchmark: which models fit in VRAM on a consumer card, how fast they run, and what they cost in watts and wall-time. The capability ranking - including these local models as their own tier - lives on the main leaderboard.

Speed × quality × tokens (3D)

drag to spin

Each glowing dot is one run. X = wall-clock time, Y = rubric score, Z = total tokens. The fast-cheap-good cluster sits in the front-bottom-right. Drag to rotate, scroll to zoom, click a point for its score breakdown.

0
evals
0
models
0
harnesses
0
hosts
0%
best score
0.0
compute (min)
0.00
M tokens