What does decisor decide?

It answers 2 kinds of question: choice picks one option from a list you give it; classify assigns the state to one or more labels. Every answer is a typed value with a probability, never free text.

How do I run decisor?

Pull the weights with systemone pull under-labs/decisor and load them with the runtime named in the manifest.

Can I use decisor commercially?

It is published under apache-2.0. Read the licence before shipping; the model card may add terms of its own.

What is decisor built on?

Its base model is qwen/qwen3.5-4b, a decisor architecture. The Model tree section shows the whole chain.

How much can decisor read at once?

The publisher has not recorded a context length. The model has 4.0B parameters.

How accurate is decisor?

Its publisher reports 73.6% decision accuracy on tc193: Under Labs' internal 193-item adaptation of LegalBench BR (10 options per item; items not released), FP8 checkpoint. Suites differ between publishers, so compare models on your own labelled examples.

decisor: open AI decision model by Under Labs

decisor-4b

Technical-preview fine-tune of Qwen3.5-4B for typed decisions in English and Brazilian Portuguese, with extra training on Brazilian legal and tax text. Picks one supplied option from one prefill pass and returns option probabilities, which are not calibrated.

FP8 weights only, to be served with Under Labs' patched SGLang image; vLLM and Transformers are untested. Scores are read from option-letter logits, like Tev1, and the model also does free-form instruction following. The maker states the probabilities are not calibrated. The SDK takes 2 to 18 options. Training data is not released. Not for consequential decisions about people. On a BF16 reference checkpoint that is not distributed, the maker reports 76.62% on JevBench's public items (a third-party benchmark, its own run).

What it decides

choice — picks one option from a set
classify — assigns a category from a fixed taxonomy

At a glance


Parameters	4B
Base model	`Qwen/Qwen3.5-4B`
Maker	Under Labs
Released	2026-10-06
License	apache-2.0
Reported accuracy	73.6%
Reported latency	49.5 decisions/s warm at concurrency 1 on an RTX PRO 6000

Get the weights

pip install systemonemodels
systemone pull under-labs/decisor

The files are served from the maker's Hugging Face repository, underlabs/decisor-4b, and verified against the checksums recorded here.

Playground

Build a request for under-labs/decisor. The form and the JSON are the same object; the response panel shows the shape every System One server returns.

Input

StateThe context the questions are asked about. Up to the model's context length.

Pick one option from a set. The answer is a probability per option and the winner.

Instructions

Options

A yes/no question. The answer is one probability that the condition holds.

Instructions

Response

Placeholder values, in the shape a run returns.

queuechoicebilling

billing
82%
security
6%
technical_support
6%
sales
6%

urgentnoul7%

This model doesn't answer live here. Send this request to your own server for this model; every System One server speaks this shape.

Call it

curl https://api.systemonemodels.ai/v1/systemone \
  -H "Authorization: Bearer $SYSTEMONE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"under-labs/decisor","state":"Subject: can't sign in after enabling 2FA\nBody: I turned on two-factor yesterday and now every code says invalid. I already reinstalled the authenticator app. Plan: Team, billed annually.","questions":{"queue":{"type":"choice","instructions":"Which team should handle this ticket?","criteria":{"billing":null,"security":null,"technical_support":null,"sales":null}},"urgent":{"type":"noul","instructions":"Is the customer locked out of a paid account right now?"}}}'

Under Labs: decisor

decisor-4b

What it decides

At a glance

Get the weights

Read more

Playground

Under Labs: decisor

decisor-4b

What it decides

At a glance

Get the weights

Read more

Playground

Providers

Pricing

Performance

Model tree

Activity

Documentation

systemone.yaml

FAQ