Drex 1.5: 9B open-weight decision model from Nace AI
A 9-billion-parameter model that scores a fixed set of typed options in one forward pass instead of generating text; open weights on Hugging Face with a hosted version on OpenRouter.
Why it matters
Matches the closed Jev decision model on the public Decision Index (58.08, within the tie band) while running on a single 24GB GPU and reading roughly 5x fewer tokens per decision.
What you could build with it
An indie developer could build a document triage service for law or procurement teams that routes each contract to the right specialist with a confidence score, using Drex's one-pass typed decisions at a fraction of a cent per request and keeping the full 128K-token document in context instead of chunking.
Does it hold up?
Too early to judge: weights went public days ago. Reported numbers (58.08 on Decision Index 0.3.1, 93.4% on 32K-128K docs) come from the vendor and one outlet; there are no independent community builds yet.
Built with Drex 1.5
- Drex deep dive on nace.ainace.ai · Official writeup: Drex 1.5 ties Jev 1.13.0 on Decision Index 0.2.1, reads a median 70 tokens per decision vs 367 for Jev, and takes up to 128K tokens in one request with 93.4% accuracy on long documents.
- MarkTechPost: Nace AI open-sources Drex 1.5marktechpost.com · Release coverage: 8.95B dense params on an NVIDIA Efficient-DLM-8B backbone, 58.08 on public Decision Index 0.3.1 (top score under 10B), bf16 on one 24GB CUDA GPU or a 9.5GB GGUF on Apple silicon.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.