TL;DR of the paper, generated automatically.

Open decision models that pick from your options, claimed to beat some big LLMs.

• What: four open models, 8B to 27B, that score options by probability.
• Benchmark: Max scores 67.52 vs 60.70 for Jev 1.14.0 on Decision Index 0.4.1.
• Speed: the paper says Max takes 183 ms, vs about 2.0 s for GPT-6 Sol.