hey, its Alti from Nace.
we released a small update with Drex Max v1.1 , both achieving the SOTA and Pareto frontier for decision models. We did all that by iterating step by step and every previous version of Drex’s open weights are available on hf.
We also released a full playbook on how we trained Drex Max in this paper: “ Towards Foundation Decision Models: Drex”
hf: https://huggingface.co/papers/2611.07854
arXiv: https://arxiv.org/pdf/2611.07854
https://preview.redd.it/7fcznstvff0i1.png?width=2000&format=png&auto=webp&s=6dc7912679a84164b6e4de9601cd67f41b5ca6a0
Drex Max v1 is weights are available here: https://huggingface.co/nace-ai/drex-max-v1 with Open RAIL-M license & v1.1 open source weights upcoming soon (we are currently optimizing for local inference).
Three cool things about Drex Max models:
it can even run on your Mac with https://github.com/nace-ai/kapitsa inference engine.
Context window is ~256k, comparing to Glide1.2(64k) and Mammoth(16k) & pplxdecider’s (256k)
Its 8x cheaper than…