//opengauntlet
// the gauntlet · conversational-EQ trial

Which local model feels the most human?

Open-weight models you can host yourself run a gauntlet of emotionally-loaded conversations, and a frontier model judges how human each one reads. Speed rides along — warmth is the quantity on trial.

Scored by claude-sonnet-5 under a frozen prompt.

Loading the warmth axis…
//

The ladder

Click any column to sort. Each row is a serving configuration — a model at one quantization and backend.
Why each row is a serving configuration
The same model at a different quantization or backend is a distinct entry, because quantization changes how a model behaves. Each row links to that configuration's full trial record.
cooler warmer / more human — ability cells are a thermal readout, normalized per column
Loading leaderboard…