JevBench v1.3.0 · one system

jeff (Logan Markewich, GLiFormer 400M)

Logan Markewich · MIT (code); GLiFormer weights per their model card · open (code and weights) · jev-rebuild

JevBench v1.3.0 score

54.4

Rank #32 of 48 ranked systems.

20.0 points behind Jev 1.13.0's 74.4.

Axes

Intelligence
46.9
Calibration
64.6
Speed
63.5
Cost
76.6

Cost

$0.0060 per 1,000 decisions (estimated).

github.com/logan-markewich/jeff

See every JevBench v1.3.0 system, method and scoring on the full leaderboard.