Drex Max v1 vs JEV-27B-VL

Drex Max v1

A 27B cousin for typed decisions over text and screenshots, several questions per request with a probability for each option. Images need an NVIDIA GPU and Nace's loader; on Apple silicon only the text backbone runs, from 24 GB. The licence isn't open source, with free commercial use only under $1 million in revenue and funding.

JEV-27B-VL

A cousin for typed decisions with probabilities over text and images, answering in one pass with optional thinking. It needs about 67 GB in BF16 on vLLM, with no AutoTrust 4-bit build or hosted API. The main risks are labels distilled from TypeSafe's Jev against its customer terms and serving through a repository script on a vLLM development build.

Drex Max v1

1. Clone github.com/nace-ai/drex-decision-models, download nace-ai/drex-max-v1 and run python -m drex_max --model weights/drex-max-v1 --request examples/request.json on an NVIDIA GPU
2. Plan for about 49 GB of GPU memory in BF16, or about 26 GB with --8bit on CUDA, before any long context
3. Don't load it with stock Transformers, vLLM or llama.cpp. Its full-attention layers must run non-causally with readout.safetensors applied, which only Nace's loader and forks do
4. On a Mac with 24 GB or more, run kapitsa-systemone --model ./drex-max-v1-affine-q4 --port 8000 --context 2048 for text-only decisions at POST /v1/systemone
5. Raise the 16,384-token default with DREX_MAX_CONTEXT=N or --max-length N, up to 262,144, and expect more memory
6. Check the licence first. Above $1 million in revenue or funding, or with a competing product, you need Nace's commercial licence

JEV-27B-VL

1. Quickest start is hf download autotrust/JEV-27B-VL then bash JEV-27B-VL/serve.sh, which serves POST /v1/decide on port 8000, but it runs the repository's serve_decide.py on a September 2026 vLLM development build
2. On a released vLLM use plain vllm serve with --enable-lora --lora-modules jev-decision=JEV-27B-VL/adapter_vllm and do the card's client-side arithmetic with decision_head.json and calibration.json
3. Keep --max-num-seqs 8, the card says larger batches return wrong System 1 probabilities on vLLM's multimodal LoRA path
4. When calling /v1/completions or /v1/chat/completions yourself pass top_k 0 and top_p 1.0, or generation_config.json's top_k 20 and top_p 0.95 truncate the probabilities
5. Choices beyond 16 options use labels the head never saw in training, and the response's adaptation field says wide-labels
6. Downscale images to about 448 px to keep vision tokens and latency down
7. Treat it as AutoTrust's model, not TypeSafe's, and its request shape differs from TypeSafe's /v1/systemone