01
The reference hosted System One model behind the Choice, Score, and Noul interface.
Official product. Access terms and pricing are controlled by TypeSafe AI. ●●●—
MODEL INDEX / PUBLIC LANDSCAPE System One
System One
Model Directory
Compare Jev with public System One alternatives, open-source reproductions, and compatibility adapters by capability, deployment, license, and primary source.
Inclusion boundary Only public projects that explicitly describe themselves as System One / Jev-like or implement a compatible typed-decision interface. Inclusion is not an endorsement.
29PUBLIC ENTRIES
24OPEN SOURCE
04ADAPTERS
04CAPABILITY AXES
01 / DIRECTORY
Compare models without mixing categories
Purpose-built models, open reproductions, and general-model adapters are separated. Capability marks describe public interfaces, not independently verified accuracy.
CAPABILITY AXIS CHOICESCORENOUL / BOOLEANRANKING
02
An open-weight System One model family for typed decisions in English and 100+ languages.
Vendor latency and benchmark figures are claims; an independent Jev comparison is linked from the source ecosystem. ●●●—
03
An open model, training recipe, and serving stack for one-step typed decisions.
The authors say it was trained independently rather than distilled from Jev. ●—●—
04
An open Thai-and-English System One model with a compatible typed-decision server.
Its model card documents known weak spots and weaker English calibration. ●●●—
05
A nano Jev replica with dynamic candidates, parallel decisions, and an end-to-end training pipeline.
Research replica; inspect its evaluation scope before comparing headline metrics. ●———
06
A tiny Jev-like decision head that can be trained and run locally on a MacBook.
Community implementation with pinned source evidence in the project radar. ●———
07
Semantic if-statements from open models, designed to run on a consumer GPU.
Independent project; not affiliated with Jev or TypeSafe AI. ●——●
08
Trains a small model to score a changing list of text options in one pass.
Research-oriented repository with Doom, chess, and Wikispeedia demonstrations. ●———
09
A non-autoregressive local decision model presented as a drop-in Jev alternative.
Latency figures are author claims and have not been independently benchmarked here. ●●●—
10
An independent Jev-compatible decision service running an open model locally.
Choice, Score, and Noul source paths are pinned in the project radar. ●●●—
11
A fine-tuned model for one-pass typed decisions with probability outputs.
System One-style community experiment; inspect the repository before production use. ●●●—
12
A small Jev / System One re-creation that maps state and typed questions to probabilities.
Independent re-creation, not an official TypeSafe AI release. ●●●—
13
An open reproduction that reads candidate logits and exposes a System One-compatible API.
Implementation evidence is pinned in the project radar. ●●●—
14
A compact decision model for agent traces, diffs, and logs with probability distributions.
Performance and calibration numbers are repository claims. ●●●—
15
A preregistered teaching experiment that reads option-letter logits instead of generating JSON.
Useful for understanding the mechanism; not presented here as production-ready. ●———
16
Jev-style parallel constrained decisions for MLX models on Apple Silicon.
A runtime technique around general models rather than a separately trained decision model. ●●●—
17
A compact non-autoregressive decision engine with calibrated uncertainty and a browser demo.
Accuracy, calibration, and latency numbers are author-reported benchmark results. ●●●—
18
An educational Jev-like visual inference experiment using shared context and direct candidate scoring.
Educational experiment, not presented as an API-compatible production model. ●——●
19
An independently implemented System One model that returns typed probability distributions in one forward pass.
Source paths and implementation claims are pinned in the project radar. ●●●—
20
Local decision-model experiments inspired by Jev and System One for NVIDIA DGX Spark.
Hardware-specific community experiment. ●●●—
21
A research project for training small multimodal models to make direct decisions.
Uses the System One label, but does not claim Jev API compatibility. ●———
22
A domain-specific open model that scores legal game actions from the screen in one forward pass.
Domain-specific game model rather than a general Jev replacement. ●———
23
A Gemma-based Jev-style decision model with a scoring head and no text generation.
Early community experiment; calibration claims are not independently verified here. ●———
24
A compact Jev-style decision model trained for Choice, Score, and Noul outputs.
Early repository; treat all performance statements as author claims. ●●●—
25
An offline, typed System One decision layer with a focus on confidence calibration.
Calibration numbers are author claims and have not been reproduced here. ●●●—
26
Adapts local language models into Jev-compatible engines using prefill-only binary inference.
Adapter technique rather than a single purpose-trained model. ●●●—
27
A wrapper that exposes Solar Mini4 through the Jev System One request shape.
Compatibility layer, not a purpose-trained System One model. ●●●—
28
Turns general open-model endpoints into a Jev-compatible classifier using candidate logprobs.
Adapter architecture; no dedicated classifier head is trained. ●——●
29
The official comparison adapter: a TypeSafeClient replacement backed by chat-model APIs.
Official adapter, but the backing chat models are not themselves System One models. ●●●—
No models match the current filters.
How to read these labels
01
Hosted model
Callable through a provider service. Jev is the reference implementation here; pricing and access belong to its provider.
02
Open implementation
Public source for local research or deployment. Open source does not automatically mean calibrated, safe, or production-ready.
03
Compatibility adapter
Wraps a general chat model in the same request shape. Useful for migration and comparison, but it does not make the backing model System One.
Found a missing model or a classification error? Submit a primary repository link for review.
Submit a model project