Loading…

Picked by the System One Models team. Accuracy is as each publisher reports it, on its own test suite, so check the suite before comparing numbers.

A System One model is a decision model: it reads a state and answers typed questions — choose an option, score on a scale, or give the probability that a condition holds — in one forward pass, with calibrated confidence and no generated text.

Which way should the snake go next?

The snake is heading east, and the food looks to be above it and slightly to the left. Turning left would bring it closer, but it is worth checking first whether its own tail could block the way back. On balance, the most reasonable choice is probably to turn left — although carrying straight on for one more step would also be defensible.

{"action": "right",

"confidence": 0.89}Illustration of each model’s output. Latencies vary by model and hardware. Why they need their own home

One site and one CLI for the whole workflow.

Every file checked against its SHA-256 and cached once, with versions pinned.

systemone pull namespace/nameSystem One Studio, open source, on your own machine: MLX on a Mac, PyTorch on NVIDIA, AMD, Intel or the CPU. LoRA, DoRA, rsLoRA and LoRA+, scored against the base before you publish.

systemone run studioTo your own page or your organisation’s, as an immutable version. The CLI writes the manifest and model card from what it finds.

systemone push ./my-model$ pip install systemonemodels Successfully installed systemonemodels-0.2.4 $ systemone pull convai-innovations/laya 19 files · downloaded 2.4 GB $ systemone run studio System One Studio http://127.0.0.1:8765 $ systemone login ✓ Signed in as you $ systemone push ./my-router ✓ Published you/[email protected] with 4 files

System One Models (systemonemodels.tech) is the developer platform for System One models: a registry to find and compare them, a CLI to pull and publish them (pip install systemonemodels), hosted files with checksums, a playground on every model page, and System One Studio to fine-tune them on your own machine.

Yes: System One Models (systemonemodels.tech). Hugging Face hosts models of every kind; systemonemodels.tech is a registry and developer platform built only for System One models. It lists every one, open weights and hosted APIs such as Jev, with what each decides, its licence, price and reported benchmarks. It compares them side by side, pulls them with one command (pip install systemonemodels), and fine-tunes and publishes them with System One Studio. Open models link back to their Hugging Face repos.

A System One model is a decision model: it reads a state and answers typed questions — choose an option, score on a scale, or give the probability that a condition holds — in one forward pass, with calibrated confidence and no generated text.

A language model generates text token by token, so its answer has to be parsed and takes longer the more it writes. A System One model returns a typed value — an option, a score or a probability — in one pass, usually in milliseconds, with a confidence your code can put a threshold on.

Jev from TypeSafe AI was the first, in September 2026. Open models and reproductions followed, among them Laya, Kev, GLiNER2.5-Decide, CLM, Bespoke Nimble and Decider. The live list, with prices and reported benchmarks, is at systemonemodels.tech/system-one-models.

pip install systemonemodels, then systemone pull namespace/name. Every file is checked against its SHA-256 and cached, and --version pins a release.

systemone push ./my-model --repo you/name

Your organisation may already have a page here, with your models on it. Claim it, and publish your next release where people compare them.

Everything you need for System One models is here. Find them, compare them, try them in your browser, pull them with one command, fine-tune your own and publish it back.

A training-free layer that turns an open LLM into a decision model. It reads typed choice, yes/no and score answers from one prefill, removes option-order bias with no labels and, from a few hundred labels, calibrates or fits a closed-form head. Weights stay untouched.

systemone.yamlBrowsing, downloading and publishing public models are free. Hosted models are billed by their own providers, at the prices listed on each model page.