Jev beats GPT-5.6 at predicting book tastes, 53x cheaper and 25x faster

venturetwins · x · 2026-09-20

The author used 1,000 Goodreads ratings as training data and held out 100 to test, asking both Jev and GPT-5.6 to guess which books would earn 5 stars.

Jev was slightly more accurate while being 53x cheaper and 25x faster — a hands-on example of small models' cost-efficiency on personalized prediction tasks.

Related event: Jev Beats GPT-5.6 in Book Taste Prediction at 1/53 the Cost(2 posts)→

Original post →

More from Models

Models channel →