Jev directory
CommunityJev Models & Reimplementations
Standard One 3B SH
Standard Thinking's 3B decision model with a joint schema head: it scores every option of every question in a scenario in one forward pass through POST /v1/systemone, and v3 self-scores 51.53 on the Decision Index 0.3 public index, up from v2's 42.90.
- Reading time
- 3 min
- Published by
- Community
- Author
- Standard Thinking
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Platforms
- API
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Decision Index 0.3 public index 51.53, self-scored with the official kit, up from v2's 42.90; the Tools area rose to 79.8 from 60.3.
- The float32 schema head reads the final layer and layer 19 and answers choice (2-255 options), noul and score questions over contexts up to 32,768 tokens.
- Decisions hidden in a 150-item long-document set improved to 85.33% from 82.00% in v2; BRIGHT slipped to 23.9 from 25.7.
- v3 adds further training on synthetic decision items with replay of earlier data; v2 had added about 2,200 hard decision items.
- Backbone: Standard One 3B v2.2 (mistral3, Ministral 3 3B text plus Pixtral vision tower); Apache-2.0; GGUF and FP8 builds are separate repositories.
Watch out
Card reviewed 2026-10-17; all Decision Index, JevBench and long-document figures are self-scored with the official kit or measured on the publisher's own sets, and the card notes that public train splits or generators of some benchmarks (HoVer among them) fed training, while evaluated items never did.
Questions about this entry
Short answers drawn from the linked project or page.
Standard One 3B with a joint schema head: it reads the scenario once, scores every option of every question in a single forward pass, and returns calibrated probabilities through the same POST /v1/systemone contract as Standard One. It does not generate text.
A letter readout limits a question to the label alphabet and needs one pass per question. The schema head attends over the hidden states of the whole request, places every question and option in one joint graph, and returns a calibrated distribution per question at once.
Choice with 2 to 255 options, noul yes/no and score questions on an ordinal scale, over contexts up to 32,768 tokens per request. The prompt format is the chat template with no default system message.
Decision Index 0.3 public index 51.53 against 42.90 for v2, area scores of 30.3 knowledge, 49.5 language, 63.3 retrieval, 79.8 tools and 36.5 arts, plus 85.33% on the long-document set. The index rows are self-scored with the official kit.
The card states the Decision Index rows come from its own runs of the public 0.3 suite, and that for some benchmarks the public train split or generator was part of training (HoVer among them), though evaluated items never were. No third-party reproduction is linked.
Standard One 3B v2.2 (mistral3: Ministral 3 3B text plus a Pixtral vision tower), fine-tuned for the head and merged into this repository under Apache-2.0. FP8 and GGUF builds are separate repositories.
Both answer typed questions with probabilities through /v1/systemone, but this model uses a learned joint schema head over hidden states instead of per-letter logits, which the card argues keeps accuracy with very many options and many questions per scenario.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
Standard Thinking's 8B decision model with a joint schema head that reads the scenario once and scores every option of every question; v4 self-scores 57.40 on the Decision Index 0.3 public index, up from v3's 48.34, with a 70.6M-parameter head.
Jev Models & Reimplementations#community#open-models#local
Community
A LoRA plus 24-slot decision head over Qwen3.8-27B from Ming Xu's train-jev project: it serves TypeSafe's System One API with 0.851 Kev transfer-v4 accuracy and ECE 0.053, and needs an 80 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
A 24 GB-GPU alternative from the train-jev project: Qwen3.5-4B plus a rank-16 LoRA and 24-slot decision head scoring 0.647 PK accuracy, 40.4 Decision Index and 43 ms p50 latency, ahead of Kev-4B at the same size.
Jev Models & Reimplementations#community#open-models#local
Community
Singularity Compute's fine-tune of Google's diffusiongemma-26B-A4B-it, a discrete-diffusion mixture-of-experts, served from an 18 GB NVFP4 checkpoint that answers choice, noul and score questions with calibrated distributions on one 48 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
Mapika's NVFP4 fine-tune of Gemma-4-31B-it answers typed Choice, Noul and Score questions with calibrated probabilities and no generated text, self-reporting a 64.27 chance-corrected public Decision Index and 39 ms median latency on a B300.
Jev Models & Reimplementations#community#open-models#local
Community
Standard Thinking's 8B LoRA adapter for the Standard One decision model: merge it with the supplied recipe and serve it, and it scores choice, noul and score labels over a scenario with calibrated probabilities via POST /v1/systemone, never generating text.
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.