Nace AI Releases Drex Decision Models with 183ms Latency and Top 3 OpenRouter Ranking

·2026.11.07 01:51

Key point

The open-source family of four models uses non-autoregressive decoding to achieve 67.52 on Decision Index 0.4.1 and rank top 3 on OpenRouter.

Nace AI released Drex, a family of four open decision models (DLM 8B, v1.5 9B, v2.0 35B MoE, Max 27B) that return probability distributions over options in a single pass without autoregressive decoding.
Drex Max achieves a 67.52 score on Decision Index 0.4.1, outperforming Jev 1.14.0 (60.70), and answers typical requests in 183 ms compared to ~2.0 s for GPT-6 Sol and ~3.5 s for Claude Opus 5.5.
Training involved full-parameter SFT on 1.15M verified records and RLAF, which improved the Decision Index by +2.0 and reduced calibration error by 0.9 pp.
Models are available in safetensors and GGUF formats with llama.cpp and Ollama support, supporting context lengths up to 262K tokens for Drex Max.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.