Open Jev: typed decision models
Open-weight replicas of TypeSafe's Jev: state + typed questions in, calibrated probabilities out, one forward pass, no decoding.
Text Classification • 0.4B • Updated • 16.7k • • 5.97kNote Most-liked open Jev. RLCD-trained, ~33 ms, pip-installable, ships an MCP extra.
convaiinnovations/laya-multilingual
Text Classification • 0.3B • Updated • 8.72k • • 427Note 100+ languages, reads up to 8k tokens.
Cloudflare/clef
Image-Text-to-Text • 27B • Updated • 77.4k • 2.37kNote Cloudflare's 27B decision model (Qwen3.8-27B base, images in the state). The one of the wave served natively by Ollama (/v1/systemone, since 0.35.1).
Cloudflare/clef-flash
Image-Text-to-Text • 9B • Updated • 34k • 929Note 9B sibling of Clef (Qwen3.5-9B base). On my 3090 via Ollama: 12.8 GB loaded, ~220 ms for three typed questions on an alert.
openjev/openjev
Zero-Shot Classification • 27B • Updated • 9.02k • 207Note Typed decisions over text, web pages and screenshots, up to 52 options.
AlexWortega/openjev
Text Classification • Updated • 654Note NLI take: decides by entailment, no per-task training (plays Doom/Minecraft in the demo).
autotrust/AutoTrust-Decider-Pro-35B
Text Classification • 36B • Updated • 1.25M • 4.11kNote Qwen3.5-35B-A3B MoE post-trained for ranking and search, running in production behind a live news recommender.
Mapika/decider-2b
Text Classification • 2B • Updated • 7.57k • 114Note Most-used of the wave; explicit open reproduction of the System One model class.
jaredpalmer/kev-4b
Text Classification • Updated • 11.1k • 181Note Family from 0.5B to 27B serving the public /v1/systemone contract.
juspay/xor
Text Generation • 35B • Updated • 387 • 56Note Industrial variant: post-trained Qwen3.6-35B-A3B, up to 255 candidates, SGLang.
togethercomputer/Tev1-4B-experimental
Text Generation • 5B • Updated • 1.82k • 39Note Together AI's Jev-inspired SFT; keeps a generative head, full recipe published.
ZefanCai/Open-Jev-9B
Updated • 54Note LoRA adapter + scalar decision head on Qwen3.5-9B, with its own server.
Hanno-Labs/bosun-v3.1-1.7b
Text Generation • Updated • 450 • 11Note Publishes DecisionBench; 84.9% vs Jev 1.13's 72.0% on it.
FrontiersMind/Lumma-fev-0.6b
Text Classification • 0.6B • Updated • 227 • 68Note Rare non-Qwen entry: fine-tuned from a base pre-trained from scratch.
MoLeMo-Lab/mojev
Text Classification • 0.9B • Updated • 16 • 4Note Only one of the wave shipping a preprint.
caiovicentino1/Eikos-27B
Text Classification • 28B • Updated • 180 • 41Note Finance and trade-finance decisions, trained with PT<->EN views.
aimeigaoshou/agent-jev
0.6B • Updated • 345 • 42Note Agent angle: predicts whether code will pass its checks, zero decoded tokens.
doofz/THX-01
Text Classification • 0.3B • Updated • 567 • 96Note Non-autoregressive mmBERT encoder (322M): ~10 ms per decision, multilingual (az/en/ru/tr/de). Smallest serious entry.
HAL-X-AI/THX-02
Zero-Shot Classification • 5B • Updated • 55.4k • 71Note Successor of THX-01: 4.5B thinking decision model, typed questions in, calibrated probabilities out, ~50 ms. Apache-2.0, ships an MCP Space.
LocalLLaMA/typed-decisions
Benchmark • Updated • 3.2k • 25.2k • 277Note Independent benchmark on the noul/choice/score schema.
Hanno-Labs/decision-bench
Viewer • Updated • 23.9k • 1.51k • 2Note DecisionBench, 23,900 frozen rows.
-
ZefanCai/Open-Jev-2B
Updated • 17 -
argos1111/modernbert-ja-310m-jev
Text Classification • 0.3B • Updated • 350 • 15 -
caiovicentino1/eikos-decisions
Viewer • Updated • 24.8k • 4.82k • 7 -
MoLeMo-Lab/mojev-mix
Viewer • Updated • 368k • 88 • 3