Skip to content

Leaderboards

Local models, real hardware, honest deltas.

Three boards: which local AI model wins which job, how the consumer-GPU stack performs at stock, and what changes when you actually open the case. Every row links to the raw data; every claim names what it isn't.

Scoring methodology: v1

Local models

Real-work benchmark of local AI models — incumbents, challengers, severity-weighted pass rates.

2 entries

Hardware (stock)

GPU and CPU benchmarks — 3DMark, Cinebench, cross-card comparisons. Stock configurations only.

12 entries

Hardware mods

Before/after measurements for thermal pad swaps, repastes, undervolts. Quality-regression checked.

1 entry

What's new

Methodology, severity weights, and raw-data locations: /leaderboards/methodology. Each section also exposes a JSON endpoint at /leaderboards/<section>.json.