Jev directory
CommunityJev Models & Reimplementations
Drex 2.0 on Fireworks AI
Nace.AI's 34.6B decision model on Fireworks AI: state plus typed yes/no, multiple-choice, or score questions return a probability for every option in one forward pass, deployable on demand with a 131K-token context window.
- Reading time
- 2 min
- Published by
- Community
- Author
- Nace.AI
- Last verified
- Oct 2026
- Platforms
- API
Highlights
- 34.6B-parameter decision model from Nace.AI, created 12 October 2026, not a mixture-of-experts model, and listed as a base model on Fireworks AI.
- Given a state and typed questions, yes/no, multiple choice, or score, it returns a probability for every option in a single forward pass.
- The blurb positions it for routing, classification, tool selection, and policy checks that need a scored answer instead of prose.
- On-demand deployments run on dedicated GPUs with Fireworks' serving stack, high reliability, and no rate limits; serverless deployment is not supported.
- The spec table lists 262k tokens while the blurb says 131K; fine-tuning, function calling, embeddings, rerankers, and image input are unsupported.
Watch out
The specification table says the model is not calibrated even though probabilities are its product, and the page contradicts itself on context length: the blurb says 131K tokens while the metadata table says 262k. No licence or weights terms are stated, serverless and fine-tuning are unsupported, and no accuracy benchmarks appear, so only on-demand GPU deployment is on offer.
Questions about this entry
Short answers drawn from the linked project or page.
A small decision model from Nace.AI hosted on Fireworks AI with the model path accounts/fireworks/models/drex-v2-0. It is built to make decisions rather than generate text, returning a probability for every option in one forward pass.
Typed questions: yes/no, multiple choice, or score. Give it a state and the questions, and it returns a probability for each option, which the page positions for routing, classification, tool selection, and policy checks.
34.6B parameters and not a mixture-of-experts model, according to the specification table. The metadata lists it as a base model created on 12 October 2026 and marked ready for deployment.
No. Serverless deployment is not supported; Drex 2.0 is available through on-demand deployments on dedicated GPUs using Fireworks' serving stack, with the page promising high reliability and no rate limits.
The page disagrees with itself: the model blurb says a 131K-token context window while the specification table lists 262k tokens. Treat the figure as unverified until Fireworks reconciles the two.
The specification table says Calibrated: No, and function calling is not supported. Embeddings, rerankers, and image input are also unsupported, and fine-tuning is unavailable, so it fits scored questions rather than tool orchestration.
The metadata names NaceXFireworks/Drex-v2.0 on Hugging Face as the model's reference. The page states no licence or weights terms, so open-weights and commercial-use questions cannot be answered from this listing.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: ASTGL replay chose local Laya routing over hosted Jev on one Mac Studio setup.
Jmeg8r15 pointsDoes an open-weight decision model beat a hosted one? Jev vs. LayaRead the thread on Hacker NewsReddit: A walkthrough of routing between models with Jev.
X: Standard One SH (StandardThinking)
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
More Jev resources like this
A CPU-first decision model from Decision Models that answers the seven questions an LLM auto-router needs (category, difficulty, stakes, four flags) in one ONNX pass; MIT weights plus a Jev-compatible hosted API.
Jev Models & Reimplementations#community#open-models#local
Community
TurinTech's independent GGUF conversion of the 27B Kev decision model for llama.cpp: the default IQ4_XS file is 13.91 GB and keeps 97.8% of the BF16 original's answers over 2,000 test decisions, with zero changed confident answers.
Jev Models & Reimplementations#community#open-models#local
Community
A 27B decision model from torchcast-ai that answers choice, noul and score questions together in one forward pass over the TypeSafe-compatible /v1/systemone format; the card reports Decision Index 0.2.1 of 65.00, +7.09 over Jev 1.13, at 33 ms per yes/no question.
Jev Models & Reimplementations#community#open-models#local
Community
Bespoke Labs' 9B decision model on Ollama, fine-tuned from Qwen3.5-9B: Apache-2.0, up to 64 typed questions per call, answers in under 100 ms on a MacBook Pro with an M5 Max, and 75.7% mean accuracy across 13 human-labelled datasets.
Jev Models & Reimplementations#community#open-models#local
Community
AutoTrust's JEV-35B page on ModelScope: metadata records a fine-tune of Qwen/Qwen3.5-35B-A3B under the Qwen3_5MoeForConditionalGeneration architecture, apache-2.0, 521 downloads, English only, with safetensors and PyTorch files.
Jev Models & Reimplementations#community#open-models#local
Community
AutoTrust's vision-language decision model on ModelScope: an adapter on Qwen3.8-27B, Apache-2.0, 1,735 downloads, with lora, safetensors, and pytorch assets, trained on the SargeDev/jev-distill-corpus-v3 corpus and English metadata.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.