Nace AI Releases Open-Weight Drex Decision Models on OpenRouter
Nace AI published four open-weight Drex decision models with inference code and OpenRouter hosting, pushing ahead of Jev 1.14 on Decision Index 0.4.1.
Open-weight competition in decision modeling accelerated in early November as Nace AI released a four-model family called Drex, publishing both model weights and inference code while making the endpoints immediately accessible through OpenRouter.
Leaderboard gains and domain specialization
Across the release—spanning drex-v1.5, drex-v2.0, drex-max, and a domain-tuned drex-max-finance variant—Nace AI adopted the standard choice, score, and noul interface with support for text, JSON, and image inputs. On the public Decision Index 0.4.1 leaderboard, Drex Max posted a score of 67.52 compared with 60.70 for TypeSafe’s Jev 1.14.0.
The inclusion of a finance-specific variant also hints at the next phase of the decision-model market. Whereas the first wave of Jev competitors targeted general-purpose triage and routing, vertical models tuned for regulated domains such as financial compliance, underwriting, and risk scoring are beginning to appear directly alongside general System One endpoints on multi-provider routers.
Self-hosting versus managed calibration
Because Nace AI ships weights and inference code openly, teams with strict data-residency requirements can deploy Drex on private GPUs while preserving API compatibility with hosted Jev. As with all leaderboard comparisons in this fast-moving category, practitioners continue to caution that calibration quality and latency under real production traffic must be verified on private evaluation sets.