Observed arrival · 2026-09-21
OWterminal’s real-silicon model board
An open-weights benchmark that ranks language-model and hardware combinations by measured token speed.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- People comparing local inference hardware and open-weight models
- Worth noticing
- Rankings use ordinal tok/s within each filter rather than a composite score.
Field notes
The board organizes local-inference observations by hardware, model, quantization or build format, runtime, and measured throughput. Its comparison frame includes consumer GPUs, Apple silicon, and two-chip DGX Spark setups rather than treating hardware as a single class. Some rows retain useful operating details—such as a 23.7 GB VRAM reading or context-length observations—while others are marked unknown or unlinked, making the dataset visibly uneven as well as practical.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue