Loading labs…
THE LABS
Companies, research groups and independent builders publishing decision models, each with its strongest releases. Open a lab to see every model it has published.
@sageasagea.space
SAGEA is a frontier AI lab from Nepal building intelligent systems for the real world. We research and engineer foundation models, reasoning systems, and production AI infrastructure.
Building System One models? Sign up and create your organisation, or claim the page we started for you.
@supersonic-labssupersoniclabs.ia.br
Brazilian lab building the Julia family of small decision models, starting with the 144M-parameter Julia 1.
@fastino-labsfastino.ai
Makers of the GLiNER2 encoder family, extended into open-weight decision models with GLiNER2.5-Decide.
@convai-innovationsconvaiinnovations.com
Ships Laya, an encoder-based decision model trained with its own RLCD recipe, and its multilingual and typed-decision checkpoints.
@startluxstartlux.com
Shanghai company focused on local AI; publishes the StartLux-Decision models in five sizes under a non-commercial licence.
@tokenrhythmtokenrhythm.ai
Beijing company working on model routing and agent post-training; publishes NeoHorse-Jev-4B on its own NeoHorse-1-4B base.
@kotoba-labsgithub.com
Publishes open-jev, a Jev-shaped DeBERTa decision model trained on public gold labels, with its full training code.
@bespoke-labsbespokelabs.ai
Data-curation company that released Nimble, an open Jev-style LoRA on Qwen3.5-9B, with its data, recipe and benchmark suite.
Bangalore payments company that post-trains and publishes Xor, a family of open decision models served through a TypeSafe-compatible API.
@mapikagithub.com
Mark Marosi's Decider family of open System One models. Unofficial listing, compiled from his public releases; the source is github.com/Mapika/decider and huggingface.co/Mapika.
Nokia's applied research team in Sunnyvale builds production agents; its researchers published AnyJev, a training-free layer that turns open LLMs into calibrated decision models (github.com/nokia-applied-research).
@iapp-technologyiapp.co.th
Thai AI company behind OpenThai-SystemOne, a Thai-and-English decision model with a 256-way slot head.
@rizzo-ai-academyrizzoaiacademy.com
Simone Rizzo's online AI academy, run by his company Inferentia SL in Spain; publishes Rizzo Flow, an open, local take on Jev.
@jared-palmerjaredpalmer.com
Trains and publishes the Kev family, LoRA adapters with a pointer head on frozen Qwen bases that serve the System One API.
@contrastive-lmcontrastive-lm.notion.site
Research group behind CLM, which scores a state against candidate actions with a contrastive objective instead of generating text.
Computer-use-agent infrastructure project publishing small Cua-S1 decision checkpoints for GUI actions.
@frontiersmindfrontiersmind.ai
Open-foundation-model company that pre-trains its own Lumma bases and publishes the Lumma-Fev decision models.
@autotrust-aiautotrust.ai
Applied research lab that adds small detachable expert blocks to frozen open models; publishes the JEV System 1 + System 2 models.
@together-aitogether.ai
Inference platform that published Tev1, a Jev-style Qwen3.5-4B fine-tune, together with the training recipe behind it.
@blockbrain-labsgithub.com
Australian lab working on trust and verification for LLMs and agents; publishes Cygnet, a typed-decision readout over frozen Gemma-4-12B-it.
Decentralised-AI company that publishes the this-that typed-decision models, fine-tuned from Mapika's decider-2b.
@akhilaaa3huggingface.co
Published Jev-Omni, a multimodal decision classifier over text, image, audio and video on Gemma 4.
@frontier-infrajebadiah.ai
Texas-based maker of the AINode local AI platform (Jason Brashear); publishes the Jebadiah open decision models in 4B, 9B and 27B.
@hit-tmggithub.com
Text-mining group at Harbin Institute of Technology, Shenzhen, known for KaLM-Embedding; publishes JevEmbed, which turns embedding models into Choice, Score and Noul deciders.
@eldanringgithub.com
Publishes Winnow, Gemma 4 fine-tunes for typed decisions that run locally through llama.cpp.
@hanno-labshannolabs.ai
Builds AI for legal, compliance and investment work (Clause Logic Inc.); publishes the Bosun calibrated typed-decision models and the DecisionBench suite.
The InternLM team at Shanghai AI Laboratory; publishes Intern-Decision, multimodal typed-decision fine-tunes of Qwen3.5 in three sizes.
@metask-labmetask.ai
Publishes Metask-Jev-4B, a multilingual, calibrated decision model trained with the Nimble objective (Hugging Face account wayfind).
@alibi-serikbaygithub.com
Distils Jev-style typed decisions into Qwen3.5 checkpoints as the JevK5 family.
@vllm-semantic-routervllm-sr.ai
The vLLM project's semantic-router team, publishing the Decision 1.0 family of open System One models and the runtime that serves them.
@meragptmeragpt.com
API platform for small single-task models; serves Decider 1, a hosted decision model that speaks the System One schema.
@standard-thinkingstandardthinking.ai
Inference provider (Standard Thinking, Inc.) that releases the Standard One decision models openly and also serves them on its Standard API.
@pngwngithub.com
Published one of the first Jev-shaped scorers, on Qwen3.5-4B, under a non-commercial licence.
@interfaze-aiinterfaze.ai
Makers of the Interfaze API, who released lev, an open Qwen3.5-4B System One model with a Jev-compatible server.
@hopit-aigithub.com
Publishes Hopper, a Qwen3.5-4B decision adapter built and measured for JevBench.
@wfzyxgithub.com
Developer of Von, a ModernBERT option-marker decision model with a Jev-compatible server.
@aac6fefhuggingface.co
Maintains the native MLX conversions of Laya for Apple silicon, the checkpoints System One Studio fine-tunes from.
@tianyu-codingsgithub.com
Author of NanoJev, a 0.6B replica trained on game environments, published with its full training pipeline.
@machinefimachinefi.com
Situated World Models for agents acting in real environments.
@typesafe-aitypesafe.ai
The company that named the System One model category and serves Jev, the first one, as a hosted decision API.
@featherless-aifeatherless.ai
Serverless inference platform and research lab; publishes SimpleJev, an open server that turns stock Hugging Face models into typed-decision endpoints.
@openaiopenai.com
Announced the Decisions API, a GPT-6 Luna endpoint that picks among developer-defined answers, at DevDay 2026.
@surogateinvergent.ai
Training stack from Invergent, a Romanian enterprise-AI company; its first decision model is Rune, a 26B-A4B model for text and images.
Enterprise AI company. djev, an open typed-decision layer for Google's DiffusionGemma, is published by David Villalón of Maisa AI, with a hosted research preview at djev.dev.
Maker of the LFM foundation models; serves d1, its first decision model, through the Liquid API.
Agent observability platform (Keywords AI, Inc.) that serves Span-01, a behaviour-scoring decision model for evals and guardrails.
@upstageupstage.ai
Korean maker of the Solar models; serves Solar Decide, a System One endpoint built on Solar Mini 4.
