System OneSystem One
All models

COMPARE

3 models, side by side.

What each one decides, how well calibrated it is, how fast it answers and what it costs to pull. Up to 4 at a time; the better value in each row is marked.

Propertymapika/decideralibi-serikbay/jevk5wfzyx/von
SummaryThe most-downloaded open System One reproduction. Merged Qwen3.5 fine-tunes, trained on about 95 public decision sets, then calibration-aware RL and a rank-64 LoRA, that softmax letter logits at an answer slot.Qwen3.5-4B with a merged rank-16 LoRA distilled from 17,408 teacher questions and 30,000 public training rows, read out as a temperature-scaled softmax over answer-letter logits. Also in 9B, 2B and GGUF.A ModernBERT-large encoder with an option-marker head. Premise and options are packed into one sequence and each option's marker is scored in a single bidirectional pass; version 1.2 makes the scoring order-invariant.
Decideschoice, score, noul, classify, routechoice, score, noul, classify, routechoice, score, noul, classify
Architecturedeciderjevk5von
Fine-tuned fromqwen/qwen3.5-2b-baseqwen/qwen3.5-4banswerdotai/modernbert-large
Licenseapache-2.0apache-2.0apache-2.0
AvailabilityOpen weightsOpen weightsOpen weights
Hosted by———
Input price———
Decision accuracy80.2%78.4%63.9%
Calibration error0.0380.0350.045
Valid action rate———
Median latency3.2 ms13.2 ms18 ms
p95 latency———
Evaluation suiteDecider 67-task regression setJevBench public hard tierJevBench public standard tier
Latest version2026.090.3.01.2.0
Variants—SHA256SUMS—
Size of latest version3.5 GB7.9 GB2.9 GB
Files482
Downloads000
Stars000
Tagssystem-one, qwen, rl, calibrated, 2bsystem-one, qwen, lora, distilled, 4bsystem-one, encoder, modernbert, 395m
UpdatedSep 25, 2026Sep 25, 2026Sep 25, 2026