Jev directory
CommunityJev Models & Reimplementations
Standard One 3B
Standard Thinking's merged BF16 3B checkpoint of Standard One: it scores choice, noul and score questions and returns probabilities through POST /v1/systemone with no generated text. The v2.2 card reports 100.00% JevBench public easy and 79.97% on many-option questions.
- Reading time
- 2 min
- Published by
- Community
- Author
- Standard Thinking
- Last verified
- Oct 2026
Highlights
- Merged BF16 3B checkpoint (base Ministral-3-3B-Instruct-2512-BF16); server code lives in the StandardOne-8B repository.
- choice, noul and score questions get probabilities in one forward pass; usage.output_tokens is 0.
- v2.2 many-option questions (53-151 options) jumped to 79.97% from 48.00%; labels now run A-Z then AA, AB, ...
- Training mixture covers English, Japanese, Chinese, Spanish, French, German, Portuguese, Russian and Korean.
- Served through SGLang 0.5.20 plus jev-adapter with --prompt-wording served --default-temperature 1.65.
Quickstart
bash
curl -s http://127.0.0.1:30120/v1/systemone -X POST -H 'content-type: application/json' -d '{
"model": "jev-latest",
"state": "Refunds require a receipt. The customer has no receipt.",
"questions": {"decision": {"type": "noul",
"instructions": "Is a refund allowed under the rule?",
"criteria": {"true": "The requirement is met.",
"false": "A requirement is missing."}}}
}'Watch out
Card reviewed 2026-10-17; the v2.1-to-v2.2 comparison table is the publisher's own served measurement (merged BF16 through SGLang 0.5.20 and jev-adapter), and the wording temperatures were fitted on three calibration suites and not refit for v2.2. Image input is supported; the published results measure text decisions.
Questions about this entry
Short answers drawn from the linked project or page.
The merged BF16 3B checkpoint of Standard Thinking's Standard One decision model, built on Ministral-3-3B-Instruct-2512-BF16. It scores a bounded set of answers for a supplied scenario and returns probabilities through POST /v1/systemone without generating free-form text.
It follows the same contract: typed questions (choice among labeled options, noul for yes/no, score for ordinal scales) about a state, with probabilities instead of decoded text. usage.output_tokens is 0 and each vector is normalized over exactly the caller's labels.
It continues training from v2.1 with additional decision data; questions with more than 26 options now use labels A-Z then AA, AB. Many-option questions (53-151 options) rose from 48.00% to 79.97% and the held-out decision set from 73.50% to 82.00%.
English, Japanese, Chinese, Spanish, French, German, Portuguese, Russian and a smaller Korean share, with performance varying by language. The card's language metadata lists en, ja, zh, es, fr, de, pt, ru and ko.
On a CUDA-capable Linux host, install SGLang 0.5.20 and the jev-adapter from the StandardOne-8B server code in separate environments, launch the engine on the merged checkpoint and start the adapter with --prompt-wording served --default-temperature 1.65.
v2.2 scores 100.00% on easy (48 items), 93.06% on standard (72) and 45.95% on hard (111). The hard proxy sits at 44.67% and the realistic transfer set at 88.17%; all figures are the publisher's own served measurements.
Apache-2.0 for the base, adapter, merged weights and shared server code. Image input is supported, but the publisher's published results measure text decisions.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Both Jev 1.13 and 1.14 score 29 of 30 on 30 Turkish buyer and seller messages
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 NewsMore Jev resources like this
The v2.2 PEFT LoRA adapter for Standard Thinking's Standard One 3B decision model, shipped with a merge recipe for Ministral-3-3B-Instruct; serving requires the merged checkpoint and the shared server code in the StandardOne-8B repository.
Jev Models & Reimplementations#community#open-models#local
Community
FRIDA-Decisions answers Russian text with structured decisions in a single encoder pass — choice, ordinal score, yes/no or ranking — and scores 0.893 on the 735-item razvilka benchmark, effectively level with TypeSafe Jev's 0.897 (McNemar p = 0.84).
Jev Models & Reimplementations#community#open-models#russian
Community
The Ollama build of iapp's OpenThai-SystemOne: a 0.8B Thai and English decision model whose v0.3 weights are fine-tuned for Ollama's /v1/systemone letter readout, reporting a 74.3 macro across 13 public subsets for Q8_0.
Jev Models & Reimplementations#community#open-models#local
Community
Standard Thinking's merged BF16 8B checkpoint of Standard One, in the repository that also carries the shared server code: bounded choice, noul and score questions get probabilities through POST /v1/systemone, with no generated answer text.
Jev Models & Reimplementations#community#open-models#multilingual
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
A 27B decision model from StartLux Labs that takes text, JSON or images up to 262,144 tokens and returns a probability per option through the TypeSafe /v1/systemone format, reporting Decision Index 0.2.1 63.88 and 208 of 231 on JevBench public.
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.