Skip to content
Live
Trending storyPast story

Hugging Face releases the Drex decision model

2 reports1 outlets9 days agoUpdate

The story so far

AI overview

Hugging Face has released the open-source decision model Drex, which returns the probability distribution of user options without autoregressive decoding, using supervised fine-tuning, RLAF, and consistency loss to improve accuracy, robustness, and latency. Its 35B parameter Max version achieved an industry-leading score of 67.52 on Decision Index 0.4.1, with response times only 9% of GPT-6 Sol's and 5% of Claude Opus 5.5's. All four models are open-source and can be obtained at https://github.com/nace-ai/drex-decision-models. Hugging Face has also released the open-source inference engine Flide, which enables multiple problems to share computation through precise state branching technology, supporting over 30 open-source decision models, and is 9.5 times faster than vLLM and SGLang in inference speed.

AI-written from the reports · Updated 8 days ago

Developments

2 developments

  1. 10 Nov, 08:00 · 1 reports
    Flide: Accelerating Decision Inference with Exact State Fork
    Hugging Face Daily Papers:Flide: Accelerating Decision Inference with Exact State Forking
  2. 6 Nov, 08:00 · 1 reports
    Towards Foundation Decision Models: Drex
    Hugging Face Daily Papers:Towards Foundation Decision Models: Drex

Coverage timeline

Follow the coverage to see every side of the story.

10 Nov
  1. Hugging Face Daily Papers
    Flide: Accelerating Decision Inference with Exact State Forking

    Flide is an open-source reasoning engine that enables multiple problems to share computation through precise state branching technology, supporting over 30 open-source decision models on Decision Index 0.4.1, with inference speed 9.5 times faster than vLLM and SGLang.

6 Nov
  1. Hugging Face Daily Papers
    Towards Foundation Decision Models: Drex

    Drex open-source decision model returns the probability distribution of user options without autoregressive decoding, using supervised fine-tuning, RLAF, and consistency loss to improve accuracy, robustness, and latency. Its 35B parameter Max version achieves an industry-leading score of 67.52 on Decision Index 0.4.1, with a response time only 9% of GPT-6 Sol and 5% of Claude Opus 5.5.

Heat over time

Not enough data yet to draw a trend.