Prism-1 Decision Sets - BEV: leaderboard

Prism-1 paper decision benchmark: single-choice decisions over 2 to 20 options read in one call without generated reasoning. BEV: 100 choice questions from each of the 12 domains of its test split (1,182 questions, 2-10 options), decisions built from public datasets and records, from page-element and tool selection to product categories, response preference, code and contract evidence. Only the models marked as published and not trained on the paper's data are published: the Jev 1.13 and Mercury Decide decision APIs and Torchcast Decision 12B (a published decision fine-tune of Gemma 4 12B). The SCQ, set-based and Prism-1 rows are methods fine-tuned on the paper's training splits and are omitted. Metric: Accuracy (%). Higher is better. Source: https://arxiv.org/html/2610.22066v1

Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.

Top models

Interactive version: theaggregate.ai/benchmark?slug=prism-1-decision-sets-bev · How It Works · Data refreshed daily, snapshot 2026-10-18.