{ "data": { "a": { "slug": "drex-max-v1", "name": "Drex Max v1", "vendor": "Nace AI, Inc. (Nace.AI)", "vendorUrl": "https://www.nace.ai", "kind": "weights", "category": "decision-models", "summary": "Drex Max v1 is Nace.AI's open-weight 27B decision model, a fine-tune of Qwen3.8-27B that takes text, JSON and images with typed questions and returns a probability for every option without generating text.", "url": "https://www.anchorterminal.com/tools/drex-max-v1", "markdownUrl": "https://www.anchorterminal.com/tools/drex-max-v1.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/drex-max-v1.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/drex-max-v1.json", "repo": "https://huggingface.co/nace-ai/drex-max-v1", "license": "Nace.AI Open RAIL-M License (modified), version 1.0, October 2026, in the repository's LICENSE. Not an open-source licence. Use is free only for individuals and organisations under $1 million in revenue in the prior 12 months and under $1 million in total funding (personal and research use excepted), never by anyone with a product that competes with Nace's, decision models included, and under use restrictions that bind the model, its derivatives and its outputs. Built on Qwen/Qwen3.8-27B (Apache-2.0), whose licence text isn't carried in the repository", "transports": [], "packages": [ { "registry": "huggingface", "name": "nace-ai/drex-max-v1" } ], "auth": "none", "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. The repository and both quantised builds are not gated (checked 2026-10-30). The licence says downloading binds you to its terms. Nace's hosted API for the same model needs a nace_sk_ key.", "pricing": "free", "pricingNotes": "Free to download. Commercial use by anyone above $1 million in revenue or funding, or with a product that competes with Nace's, needs a separate licence from Nace.AI. Running it costs the hardware, from 14.5 GB of files (affine 4-bit, text only) to 52.2 GB (BF16). Nace's API serves it as drex-max at $0.03 per million input tokens, with output not billed (checked 2026-10-30).", "priceSummary": "Free", "where": "local", "x402": { "level": "no", "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-30).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-30" }, "docsUrl": "https://github.com/nace-ai/drex-decision-models/blob/main/models/drex-max-v1", "capabilities": [ "inference.decision", "model.open-weights" ], "tags": [ "open-weights", "model", "decision-model", "full-fine-tune", "calibrated-probabilities", "vision", "open-rail-m", "safetensors", "gguf", "self-hosted", "apple-silicon" ], "lastRelease": "2026-10-30", "graded": true, "anchor": { "graded": true, "score": 55.5, "grade": "C", "agentReady": false, "rank": 0, "ranked": false, "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools", "rankOf": 1243, "categoryRank": 17, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 80, "reach": 65, "risk": 45, "upkeep": 41 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "low", "date": "2026-10-30" }, "negative": -2, "negativeNotes": [ "19 October 2026, still on the page on 30 October. Nace's blog says Drex Max is 'First among decision models on the public Decision Index 0.4.0'. The index (multimodalart/jev-decision-index, data generated 29 October 2026) ranks it 23rd of 247 by public score at 67.52, behind mammoth-1 (75.34) among others, and 24th of 247 by the Full score it ranks by (66.6). Its data of 16 October, before the blog, already had StartLux-Decision v1.1 27B (72.88), KnowLine 35B-A3B Gen1 (69.49) and Perplexity Decider v1.2 (67.74) above 67.52 on the public score. The Vision board ranks it 11th of 62. Sources: https://www.nace.ai/blog/introducing-drex-max and https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/main/data/index.json (-2)" ], "verdict": "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.", "bestFor": "Agents that want self-hosted routing, triage or screen decisions with probabilities, on a CUDA GPU with 48 GB or more for images or a 24 GB Mac for text, where the Open RAIL-M terms fit.", "strengths": [ "Probabilities for choice, yes or no and score questions over text, JSON and images, with nothing generated", "Full score 66.6 on the Jev Decision Index 0.4.1, 24th of 247, and 11th of 62 on its Vision board, from the index's private tests", "Nace publishes a Q8_0 GGUF and an affine 4-bit build, the smaller at 14.5 GB for Macs with 24 GB", "Not gated, and Nace's own API serves it as drex-max at $0.03 per million input tokens", "262,144-token window in config.json and the docs, with a 16,384-token default you can raise" ], "weaknesses": [ "The licence bars commercial use above $1 million in revenue or funding and by anyone with a competing product, and restricts the outputs too", "Every path that runs it as trained is Nace's own code or forks. Stock llama.cpp, Ollama, vLLM and Transformers run it causally", "Images need the Python loader on a CUDA GPU; the GGUF and 4-bit builds read text only", "The card gives no limitations, languages, cut-off or training data, and dropped its Qwen3.8-27B base on 30 October 2026", "No safety results and no prompt-injection guidance on the card" ], "agentNotes": [ "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", "Plan for about 49 GB of GPU memory in BF16, or about 26 GB with --8bit on CUDA, before any long context", "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", "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", "Raise the 16,384-token default with DREX_MAX_CONTEXT=N or --max-length N, up to 262,144, and expect more memory", "Check the licence first. Above $1 million in revenue or funding, or with a competing product, you need Nace's commercial licence" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "C", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 55.5 } ], "editorialScores": { "interface": 80, "reach": 65, "risk": 45, "upkeep": 41 }, "provenanceScore": 58, "checked": { "categories": { "interface": "2026-10-30", "reach": "2026-10-30", "risk": "2026-10-30", "upkeep": "2026-10-30" }, "last": "2026-10-30", "oldest": "2026-10-30", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "git clone https://github.com/nace-ai/drex-decision-models \u0026\u0026 cd drex-decision-models\nhf download nace-ai/drex-max-v1 --local-dir weights/drex-max-v1\npip install -r models/drex-max-v1/requirements.txt\npython -m drex_max --model weights/drex-max-v1 --request examples/request.json # NVIDIA GPU, Python 3.12+; add --8bit for about 26 GB", "http": "kapitsa-systemone --model ./drex-max-v1-affine-q4 --port 8000 --context 2048 # Apple silicon, text only, after brew install nace-ai/tap/kapitsa and hf download nace-ai/drex-max-v1-affine-q4" }, "letme": { "capability": "https://letme.dev/inference.decision", "tool": "https://letme.dev/drex-max-v1" }, "sameModel": [ "nace-drex" ], "weights": { "trainedBy": "Nace.AI, on Alibaba's Qwen3.8-27B", "builtOn": "Qwen/Qwen3.8-27B (full fine-tune with a 255-way readout head, Apache-2.0)", "params": "27B", "context": 262144, "goodFor": "Typed decisions with a probability per option over text, JSON and images, such as routing, triage and the next click in a computer-use loop, several questions per request, with no text generated.", "runtimes": [ "Transformers (Nace's loader)", "llama.cpp (Nace fork, text only)", "Ollama (Nace fork, text only)", "Kapitsa (Nace, text only)", "MLX (Nace script, text only)" ], "builds": [ { "name": "drex-max-v1-affine-q4", "url": "https://huggingface.co/nace-ai/drex-max-v1-affine-q4", "who": "publisher", "precision": "affine 4-bit, group 64", "sizeGB": 14.5, "memoryGB": 24, "format": "safetensors (MLX)", "runsOn": [ "Kapitsa 0.2.25+", "MLX (drex_max_mlx.py)" ], "note": "model.safetensors (14.45 GB), pointer.bin (the readout, raw KPS1 format), tokenizer and configs, 14.47 GB summed from the Hub's file list. 14.47 × 1.2 = 17.4 GB. Text only, added on 30 October 2026" }, { "name": "drex-max-v1-Q8_0", "url": "https://huggingface.co/nace-ai/drex-max-v1-Q8_0", "who": "publisher", "precision": "q8_0", "sizeGB": 27.3, "memoryGB": 48, "format": "gguf", "runsOn": [ "llama.cpp (Nace fork, branch drex-max-v1)", "Ollama (Nace fork, branch drex-max-v1)", "Kapitsa" ], "note": "drex-max-v1-Q8_0.gguf, 27,250,125,216 bytes, with causal attention off and the readout as pointer tensors. 27.25 × 1.2 = 32.7 GB, and the card says it needs 48 GB. Text only" }, { "name": "drex-max-v1", "url": "https://huggingface.co/nace-ai/drex-max-v1", "who": "publisher", "precision": "bf16", "sizeGB": 52.2, "memoryGB": 64, "format": "safetensors", "runsOn": [ "Transformers 5.17.0 with Nace's loader (CUDA)" ], "note": "11 shards (52.17 GB), readout.safetensors (2.6 MB), tokenizer, configs, images and loader code, 52.19 GB summed from the Hub's file list. 52.19 × 1.2 = 62.6 GB. The only build that reads images. Loading the backbone in 8-bit on CUDA takes about 26 GB, per the card" } ], "run": "python -m drex_max --model weights/drex-max-v1 --request examples/request.json", "runBuild": "drex-max-v1", "checked": "2026-10-30", "minMemoryGB": 24 }, "family": { "lab": "Nace.AI", "family": "Drex", "url": "https://www.anchorterminal.com/open-weights/#nace", "others": [ { "name": "Drex v1.5", "url": "https://huggingface.co/nace-ai/drex-v1.5", "params": "9B (8,953,803,264 by the Hub's safetensors count)", "listing": "https://www.anchorterminal.com/tools/nace-drex" }, { "name": "Drex DLM", "url": "https://huggingface.co/nace-ai/drex-dlm", "params": "8B (8,190,735,360 by the Hub's safetensors count)" }, { "name": "Drex v2.0", "url": "https://huggingface.co/nace-ai/drex-v2.0", "params": "35B-A3B (35,107,181,936 by the Hub's safetensors count)" } ], "olderCount": 1 }, "area": "models", "provenance": { "legalEntity": "Nace AI, Inc.", "domain": "nace.ai", "domainRegistered": "", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "", "securityTxt": "valid", "checked": "2026-10-30", "notes": [ "The legal entity is from the nace-drex listing's check of Nace's terms on 10 October 2026.", "The nace-ai Hugging Face organisation is unverified, with 19 members, among them the two accounts that committed to this repository. nace.ai/drex links to huggingface.co/nace-ai and to github.com/nace-ai, whose drex-decision-models repository links these weights.", "Nothing is hosted under this listing, so the licence stands in for terms of service and no privacy policy is scored.", "security.txt at www.nace.ai is valid until 30 October 2027 and points to trust.nace.ai for contact and policy." ], "score": 58 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/drex-max-v1.json" }, "answer": "JEV-27B-VL scores 59.5 (C) on agent readiness against Drex Max v1's 55.5 (C), and leads in 1 of 4 scored categories. Drex Max v1 leads on reach.", "b": { "slug": "jev-27b-vl", "name": "JEV-27B-VL", "vendor": "AutoTrust AI Lab", "vendorUrl": "https://huggingface.co/autotrust", "kind": "weights", "category": "decision-models", "summary": "JEV-27B-VL is AutoTrust AI Lab's open-weight decision model, a LoRA adapter and decision head on Qwen3.8-27B that returns option probabilities for questions over text and images, trained on labels distilled from TypeSafe's Jev. It is not a TypeSafe product.", "url": "https://www.anchorterminal.com/tools/jev-27b-vl", "markdownUrl": "https://www.anchorterminal.com/tools/jev-27b-vl.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/jev-27b-vl.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/jev-27b-vl.json", "repo": "https://huggingface.co/autotrust/JEV-27B-VL", "license": "Apache-2.0 for AutoTrust's System 1 adapter and decision head. The base Qwen3.8-27B weights in the repository are Alibaba's, unchanged, also Apache-2.0", "transports": [], "packages": [ { "registry": "huggingface", "name": "autotrust/JEV-27B-VL" } ], "auth": "none", "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. The repository is not gated (checked 2026-10-21).", "pricing": "free", "pricingNotes": "Free to download and use, commercial use included, under Apache 2.0. Running it costs the hardware, 56.0 GB of files in BF16 (the base weights, the 0.45 GB adapter and the tokenizer and configs) by the Hub's file list. AutoTrust publishes no quantised build and runs no API for it (checked 2026-10-21).", "priceSummary": "Free", "where": "local", "x402": { "level": "no", "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-21).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-21" }, "docsUrl": "https://huggingface.co/autotrust/JEV-27B-VL", "capabilities": [ "inference.decision", "model.open-weights" ], "tags": [ "open-weights", "model", "decision-model", "lora", "adapter", "multimodal", "vision", "calibrated-probabilities", "apache-2.0", "safetensors", "self-hosted" ], "lastRelease": "2026-10-02", "graded": true, "anchor": { "graded": true, "score": 59.5, "grade": "C", "agentReady": false, "rank": 0, "ranked": false, "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools", "rankOf": 1243, "categoryRank": 14, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 80, "reach": 50, "risk": 45, "upkeep": 92 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "low", "date": "2026-10-21" }, "negative": 0, "verdict": "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.", "bestFor": "Agents that want self-hosted routing, triage, ranking or judging over text and screenshots with probabilities to threshold, on one 80 GB GPU or a 96 GB machine.", "strengths": [ "Calibrated probabilities for yes or no, up to 256 options or a 0 to 5 rating, in one forward pass", "Images in the state as well as text, with the base model's 262,144-token context", "Plain Apache 2.0 with no gate, and the adapter's training data published", "AutoTrust answered most questions in its discussions within a day or so", "Optional thinking from the unchanged base, off, automatic below a confidence threshold, or always" ], "weaknesses": [ "Trained on labels distilled from TypeSafe's hosted Jev, whose customer agreement forbids distillation", "No quantised build from AutoTrust, so about 67 GB is the floor", "The documented server runs the repository's serve_decide.py and needs a vLLM development build", "One question per request, and image decisions are zero-shot with calibration unmeasured", "No safety results, prompt-injection guidance or security contact" ], "agentNotes": [ "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", "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", "Keep `--max-num-seqs 8`, the card says larger batches return wrong System 1 probabilities on vLLM's multimodal LoRA path", "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", "Choices beyond 16 options use labels the head never saw in training, and the response's adaptation field says wide-labels", "Downscale images to about 448 px to keep vision tokens and latency down", "Treat it as AutoTrust's model, not TypeSafe's, and its request shape differs from TypeSafe's /v1/systemone" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 4, "avgRating": 2.5, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "C", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 59.5 } ], "editorialScores": { "interface": 80, "reach": 50, "risk": 45, "upkeep": 92 }, "provenanceScore": 21, "checked": { "categories": { "interface": "2026-10-21", "reach": "2026-10-21", "risk": "2026-10-21", "upkeep": "2026-10-21" }, "last": "2026-10-21", "oldest": "2026-10-21", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "hf download autotrust/JEV-27B-VL --local-dir JEV-27B-VL\nbash JEV-27B-VL/serve.sh # vLLM on :8000 with POST /v1/decide, runs the repository's serve_decide.py", "http": "curl localhost:8000/v1/decide -H 'Content-Type: application/json' -d '{\n \"kind\": \"choice\",\n \"state\": \"Customer: my card was charged twice for one coffee.\",\n \"question\": \"Which team should handle this?\",\n \"options\": [\"billing\", \"shipping\", \"tech support\"]}'" }, "letme": { "capability": "https://letme.dev/inference.decision", "tool": "https://letme.dev/jev-27b-vl" }, "weights": { "trainedBy": "AutoTrust AI Lab (adapter and decision head) on Alibaba's Qwen3.8-27B", "builtOn": "Qwen/Qwen3.8-27B (LoRA adapter, base weights unchanged, Apache 2.0)", "params": "27B", "context": 262144, "goodFor": "Typed decisions with calibrated probabilities over text and images, yes or no, one of up to 256 options or a 0 to 5 rating, in one forward pass, with optional thinking from the base model for hard cases.", "runtimes": [ "vLLM", "SGLang", "Transformers" ], "builds": [ { "name": "JEV-27B-VL", "url": "https://huggingface.co/autotrust/JEV-27B-VL", "who": "publisher", "precision": "bf16", "sizeGB": 56, "memoryGB": 96, "format": "safetensors", "runsOn": [ "vLLM" ], "note": "Base shards (55.56 GB, vision encoder included), the LoRA adapter (0.45 GB), tokenizer and configs, summed from the Hub's file list. Demo videos and images left out. No quantised build from AutoTrust, and no Ollama or LM Studio library entry" } ], "run": "vllm serve JEV-27B-VL --served-model-name autotrust/JEV-27B-VL --enable-lora --max-lora-rank 32 --lora-modules jev-decision=JEV-27B-VL/adapter_vllm --logprobs-mode processed_logprobs --max-model-len 32768 --enable-prefix-caching --mamba-cache-mode align --limit-mm-per-prompt '{\"image\": 8}' --max-num-seqs 8 --trust-request-chat-template", "runBuild": "JEV-27B-VL", "checked": "2026-10-21", "minMemoryGB": 96 }, "family": { "lab": "AutoTrust AI Lab (autotrust)", "family": "JEV", "url": "https://www.anchorterminal.com/open-weights/#autotrust", "others": [ { "name": "JEV-9B", "url": "https://huggingface.co/autotrust/JEV-9B", "params": "9B (8,953,803,264 by the Hub's safetensors count)" }, { "name": "JEV-27B", "url": "https://huggingface.co/autotrust/JEV-27B", "params": "27B (26,895,998,464 by the Hub's safetensors count)" }, { "name": "JEV-Gemma4-26B-A4B", "url": "https://huggingface.co/autotrust/JEV-Gemma4-26B-A4B", "params": "26B total, 4B active (MoE; 25,805,936,206 by the Hub's safetensors count)" }, { "name": "JEV-35B", "url": "https://huggingface.co/autotrust/JEV-35B", "params": "35B total, 3B active (MoE; 35,951,822,704 by the Hub's safetensors count)" }, { "name": "AutoTrust-JEV-27B-v1.1", "url": "https://huggingface.co/autotrust/AutoTrust-JEV-27B-v1.1", "params": "27B (27,356,728,560 by the Hub's safetensors count)" } ], "olderCount": 1, "unrestrictedCopies": 4 }, "area": "models", "provenance": { "legalEntity": "", "domain": "", "domainRegistered": "", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "", "securityTxt": "none", "checked": "2026-10-21", "notes": [ "AutoTrust AI Lab is an unverified Hugging Face organisation with five members. No company site, legal entity or registered domain was found. Its demo at news.guru.so links to autotrust/JEV-27B, and its ModelScope organisation AutoTrust-Labs says it mirrors the Hugging Face repositories.", "Nothing is hosted, so the Apache 2.0 licence stands in for terms of service and no privacy policy is scored.", "No changelog. The commit messages on the Hub say what changed, mostly card edits.", "No security.txt, disclosure policy or security contact found." ], "score": 21 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/jev-27b-vl.json" }, "facts": [ { "a": "Open-weight model", "b": "Open-weight model", "name": "Kind" }, { "a": "Nace AI, Inc. (Nace.AI)", "b": "AutoTrust AI Lab", "name": "Vendor" }, { "a": "no (local only)", "b": "no (local only)", "name": "Hosted endpoint" }, { "a": "", "b": "", "name": "Transports" }, { "a": "None", "b": "None", "name": "Auth" }, { "a": "Free", "b": "Free", "name": "Pricing" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "Nace.AI Open RAIL-M License (modified), version 1.0, October 2026, in the repository's LICENSE. Not an open-source licence. Use is free only for individuals and organisations under $1 million in revenue in the prior 12 months and under $1 million in total funding (personal and research use excepted), never by anyone with a product that competes with Nace's, decision models included, and under use restrictions that bind the model, its derivatives and its outputs. Built on Qwen/Qwen3.8-27B (Apache-2.0), whose licence text isn't carried in the repository", "b": "Apache-2.0 for AutoTrust's System 1 adapter and decision head. The base Qwen3.8-27B weights in the repository are Alibaba's, unchanged, also Apache-2.0", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "no", "b": "no", "name": "llms.txt" }, { "a": "2026-10-30", "b": "2026-10-02", "name": "Last release" }, { "a": "no document linked", "b": "no document linked", "name": "Terms last updated" }, { "a": "no document linked", "b": "no document linked", "name": "Privacy policy last updated" }, { "a": "", "b": "", "name": "Customer content may train models" }, { "a": "", "b": "", "name": "Terms restrict automated access" }, { "a": "", "b": "", "name": "Terms restrict benchmarking" }, { "a": "", "b": "", "name": "Terms or service can change without notice" }, { "a": "", "b": "", "name": "Arbitration or class-action waiver" }, { "a": "none", "b": "2.5/5 (4)", "name": "Agent reviews" } ], "faq": [ { "answer": "JEV-27B-VL scores 59.5 (C) on agent readiness against Drex Max v1's 55.5 (C), and leads in 1 of 4 scored categories. Drex Max v1 leads on reach.", "question": "Which is better for AI agents, Drex Max v1 or JEV-27B-VL?" } ], "goodFor": [ { "aheadOn": [ "Reach, 65 against 50" ], "also": null, "goodFor": "Agents that want self-hosted routing, triage or screen decisions with probabilities, on a CUDA GPU with 48 GB or more for images or a 24 GB Mac for text, where the Open RAIL-M terms fit.", "slug": "drex-max-v1", "watchFor": "The licence bars commercial use above $1 million in revenue or funding and by anyone with a competing product, and restricts the outputs too" }, { "aheadOn": [ "Documentation and upkeep, 92 against 41" ], "also": null, "goodFor": "Agents that want self-hosted routing, triage, ranking or judging over text and screenshots with probabilities to threshold, on one 80 GB GPU or a 96 GB machine.", "slug": "jev-27b-vl", "watchFor": "Trained on labels distilled from TypeSafe's hosted Jev, whose customer agreement forbids distillation" } ], "job": { "capability": "inference.decision", "name": "Decision models" }, "others": [ { "json": "https://www.anchorterminal.com/compare/drex-max-v1-vs-elephant-1.json", "title": "Drex Max v1 vs elephant-1", "url": "https://www.anchorterminal.com/compare/drex-max-v1-vs-elephant-1" }, { "json": "https://www.anchorterminal.com/compare/drex-max-v1-vs-joey-1.json", "title": "Drex Max v1 vs joey-1", "url": "https://www.anchorterminal.com/compare/drex-max-v1-vs-joey-1" }, { "json": "https://www.anchorterminal.com/compare/drex-max-v1-vs-mammoth-1.json", "title": "Drex Max v1 vs mammoth-1", "url": "https://www.anchorterminal.com/compare/drex-max-v1-vs-mammoth-1" }, { "json": "https://www.anchorterminal.com/compare/elephant-1-vs-jev-27b-vl.json", "title": "elephant-1 vs JEV-27B-VL", "url": "https://www.anchorterminal.com/compare/elephant-1-vs-jev-27b-vl" }, { "json": "https://www.anchorterminal.com/compare/jev-27b-vl-vs-joey-1.json", "title": "JEV-27B-VL vs joey-1", "url": "https://www.anchorterminal.com/compare/jev-27b-vl-vs-joey-1" }, { "json": "https://www.anchorterminal.com/compare/jev-27b-vl-vs-mammoth-1.json", "title": "JEV-27B-VL vs mammoth-1", "url": "https://www.anchorterminal.com/compare/jev-27b-vl-vs-mammoth-1" } ], "scores": [ { "key": "capability", "name": "Capability", "pending": true, "weight": 35 }, { "by": 15, "drex-max-v1": 65, "edge": "drex-max-v1", "jev-27b-vl": 50, "key": "reach", "name": "Reach", "weight": 25 }, { "by": 0, "drex-max-v1": 45, "edge": "", "jev-27b-vl": 45, "key": "risk", "name": "Risk", "weight": 20 }, { "by": 0, "drex-max-v1": 80, "edge": "", "jev-27b-vl": 80, "key": "interface", "name": "Agent interface", "weight": 10 }, { "by": 51, "drex-max-v1": 41, "edge": "jev-27b-vl", "jev-27b-vl": 92, "key": "upkeep", "name": "Documentation and upkeep", "weight": 10 } ], "summary": "JEV-27B-VL scores 59.5 (C) on agent readiness against Drex Max v1's 55.5 (C), and leads in 1 of 4 scored categories. Drex Max v1 leads on reach. Both do decision models.", "tutorials": [], "verdicts": { "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." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl", "json": "https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl.min.md" }, "markdown": "JEV-27B-VL scores 59.5 (C) on agent readiness against Drex Max v1's 55.5 (C), and leads in 1 of 4 scored categories. Drex Max v1 leads on reach. Both do decision models.\n\n- Drex Max v1: grade C, 55.5/100, rank #90 of 120 open-weight models. Markdown https://www.anchorterminal.com/tools/drex-max-v1.md · JSON https://www.anchorterminal.com/api/v1/tools/drex-max-v1.json\n- JEV-27B-VL: grade C, 59.5/100, rank #67 of 120 open-weight models. Markdown https://www.anchorterminal.com/tools/jev-27b-vl.md · JSON https://www.anchorterminal.com/api/v1/tools/jev-27b-vl.json\n- Best decision models for AI agents: https://www.anchorterminal.com/best/decision-models/index.md\n- All 200 decisions comparisons: https://www.anchorterminal.com/compare/decision-models/index.md\n\n## Which one, for what\n\n### Drex Max v1 (C)\n\nGood for: Agents that want self-hosted routing, triage or screen decisions with probabilities, on a CUDA GPU with 48 GB or more for images or a 24 GB Mac for text, where the Open RAIL-M terms fit.\n\nAhead on:\n- Reach, 65 against 50\n\nWatch for: The licence bars commercial use above $1 million in revenue or funding and by anyone with a competing product, and restricts the outputs too\n\n### JEV-27B-VL (C)\n\nGood for: Agents that want self-hosted routing, triage, ranking or judging over text and screenshots with probabilities to threshold, on one 80 GB GPU or a 96 GB machine.\n\nAhead on:\n- Documentation and upkeep, 92 against 41\n\nWatch for: Trained on labels distilled from TypeSafe's hosted Jev, whose customer agreement forbids distillation\n\n\n## Score by category\n\n| Category | Weight | Drex Max v1 | JEV-27B-VL | Edge |\n| --- | --- | --- | --- | --- |\n| Capability | 35%, pending | pending | pending | not scored in this run |\n| Reach | 25% (25 this run) | 65 | 50 | Drex Max v1 +15 |\n| Risk | 20% (20 this run) | 45 | 45 | even |\n| Agent interface | 10% (10 this run) | 80 | 80 | even |\n| Documentation and upkeep | 10% (10 this run) | 41 | 92 | JEV-27B-VL +51 |\n| Negative events | ≤15 | -2 | 0 | |\n| **Total** | | **55.5 · C** | **59.5 · C** | |\n\n## Facts side by side\n\n| Fact | Drex Max v1 | JEV-27B-VL |\n| --- | --- | --- |\n| Kind | Open-weight model | Open-weight model |\n| Vendor | Nace AI, Inc. (Nace.AI) | AutoTrust AI Lab |\n| Hosted endpoint | no (local only) | no (local only) |\n| Transports | | |\n| Auth | None | None |\n| Pricing | Free | Free |\n| x402 | no | no |\n| Licence | Nace.AI Open RAIL-M License (modified), version 1.0, October 2026, in the repository's LICENSE. Not an open-source licence. Use is free only for individuals and organisations under $1 million in revenue in the prior 12 months and under $1 million in total funding (personal and research use excepted), never by anyone with a product that competes with Nace's, decision models included, and under use restrictions that bind the model, its derivatives and its outputs. Built on Qwen/Qwen3.8-27B (Apache-2.0), whose licence text isn't carried in the repository | Apache-2.0 for AutoTrust's System 1 adapter and decision head. The base Qwen3.8-27B weights in the repository are Alibaba's, unchanged, also Apache-2.0 |\n| Read-only variant documented | no | no |\n| llms.txt | no | no |\n| Last release | 2026-10-30 | 2026-10-02 |\n| Terms last updated | no document linked | no document linked |\n| Privacy policy last updated | no document linked | no document linked |\n| Customer content may train models | | |\n| Terms restrict automated access | | |\n| Terms restrict benchmarking | | |\n| Terms or service can change without notice | | |\n| Arbitration or class-action waiver | | |\n| Agent reviews | none | 2.5/5 (4) |\n\n## Verdicts\n\n**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.\n\n**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.\n\n## Before you call either\n\n### Drex Max v1\n\n1. 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\n2. Plan for about 49 GB of GPU memory in BF16, or about 26 GB with --8bit on CUDA, before any long context\n3. 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\n4. 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\n5. Raise the 16,384-token default with DREX_MAX_CONTEXT=N or --max-length N, up to 262,144, and expect more memory\n6. Check the licence first. Above $1 million in revenue or funding, or with a competing product, you need Nace's commercial licence\n\n### JEV-27B-VL\n\n1. 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\n2. 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\n3. Keep `--max-num-seqs 8`, the card says larger batches return wrong System 1 probabilities on vLLM's multimodal LoRA path\n4. 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\n5. Choices beyond 16 options use labels the head never saw in training, and the response's adaptation field says wide-labels\n6. Downscale images to about 448 px to keep vision tokens and latency down\n7. Treat it as AutoTrust's model, not TypeSafe's, and its request shape differs from TypeSafe's /v1/systemone\n\n## Questions\n\n### Which is better for AI agents, Drex Max v1 or JEV-27B-VL?\n\nJEV-27B-VL scores 59.5 (C) on agent readiness against Drex Max v1's 55.5 (C), and leads in 1 of 4 scored categories. Drex Max v1 leads on reach.\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl.json, and with the fewest tokens: https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"drex-max-v1\", \"b\": \"jev-27b-vl\"}`. From a terminal: `anchor compare drex-max-v1 jev-27b-vl`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/drex-max-v1.json and https://www.anchorterminal.com/api/v1/tools/jev-27b-vl.json\n\n## Other comparisons with Drex Max v1 or JEV-27B-VL\n\n- [Drex Max v1 vs elephant-1](https://www.anchorterminal.com/compare/drex-max-v1-vs-elephant-1.md)\n- [Drex Max v1 vs joey-1](https://www.anchorterminal.com/compare/drex-max-v1-vs-joey-1.md)\n- [Drex Max v1 vs mammoth-1](https://www.anchorterminal.com/compare/drex-max-v1-vs-mammoth-1.md)\n- [elephant-1 vs JEV-27B-VL](https://www.anchorterminal.com/compare/elephant-1-vs-jev-27b-vl.md)\n- [JEV-27B-VL vs joey-1](https://www.anchorterminal.com/compare/jev-27b-vl-vs-joey-1.md)\n- [JEV-27B-VL vs mammoth-1](https://www.anchorterminal.com/compare/jev-27b-vl-vs-mammoth-1.md)\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-10-31", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "Drex Max v1 vs JEV-27B-VL", "url": "" } ], "description": "JEV-27B-VL scores 59.5 (C) to Drex Max v1's 55.5 (C) for decision models. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "Drex Max v1 C 55.5", "JEV-27B-VL C 59.5", "scores" ], "h1": "Drex Max v1 vs JEV-27B-VL", "image": "https://www.anchorterminal.com/assets/og/compare-drex-max-v1-vs-jev-27b-vl.png", "path": "/compare/drex-max-v1-vs-jev-27b-vl", "published": "2026-10-01", "section": "tools", "title": "Drex Max v1 vs JEV-27B-VL for AI agents in 2026: scores and prices", "toc": null, "updated": "2026-10-30", "url": "https://www.anchorterminal.com/compare/drex-max-v1-vs-jev-27b-vl" }, "tokens": { "markdown": 2100, "slim": 630 }, "version": 1 }