---
format: "aidr-story-markdown/v1"
id: "e025099f0efd5e9f2f4f215285263f491689eeae1d83b604ff2c45d4552e6593"
canonical_url: "https://aidr.today/e025099f?lang=en"
title: "Towards Foundation Decision Models: Drex"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T00:00:00.000Z"
category: "Models"
topics: ["llm"]
source_urls: ["https://huggingface.co/papers/2611.07854"]
summary: "Researchers have introduced Drex, a family of four open decision models with 8B, 9B, 35B, and 27B parameters designed to return probability distributions over user-supplied options without autoregressive decoding. The models are trained using supervised fine-tuning, reinforcement learning from adjusted feedback, and vision tuning for the Max variant, with ablation studies showing improvements in accuracy and robustness."
---

# Towards Foundation Decision Models: Drex

> [Open the canonical story](<https://aidr.today/e025099f?lang=en>)

**Published:** 2026-11-06T00:00:00.000Z
**Category:** Models
**Topics:** llm

## Summary

Researchers have introduced Drex, a family of four open decision models with 8B, 9B, 35B, and 27B parameters designed to return probability distributions over user\-supplied options without autoregressive decoding\. The models are trained using supervised fine\-tuning, reinforcement learning from adjusted feedback, and vision tuning for the Max variant, with ablation studies showing improvements in accuracy and robustness\.

## Sources

- [Story source](<https://huggingface.co/papers/2611.07854>)