Jev beats GPT-5.6 at predicting reading taste, 53x cheaper and 25x faster
venturetwins · x · 2026-09-20
Using 1,000 Goodreads ratings as training data and holding out 100 for testing, the author had Jev and GPT-5.6 predict which books would earn 5 stars. Jev was slightly more accurate while costing 53x less and running 25x faster — surprisingly without needing a deeper reasoning model. The author argues price and speed are the difference between a consumer app that works and one that's infeasible, and is excited to see what apps Jev unlocks.
Related event: Jev Beats GPT-5.6 in Book Taste Prediction at 1/53 the Cost(2 posts)→
More from Models
- Pangram extension shows AI-flagged X posts come from outside the AI field — dylfreed · 2026-09-20
- trycua open-sources CUA-S1-FORMS, first of small specialized computer-use model family — altryne · 2026-09-20
- Fine-tune a calibrated LLM classifier for $2: most classification tasks don't need frontier models — bingxu_ · 2026-09-20
- Voyage AI founder Tengyu Ma: frontier models at Ultra settings are still fundamentally dumb — lateinteraction · 2026-09-20
- Users Are Leaving Gemini for Claude/Codex — How Big a Lead Would Bring Them Back? — Helpful-Collar7159 · 2026-09-20
- OpenAI 20x plan user burns 5% of weekly quota in one hour: "new low" — No_Twist_678 · 2026-09-20