mogan-decision-31B-it
Mogan AI's 31B decision model: a LoRA adapter (r=128) on Gemma-4-31B-it that answers choice (2 to 255 options) and noul questions in one forward pass, reading a probability for every option from the option-label logits. Weights are non-commercial.
Adapter only; it needs Gemma-4-31B-it and about 70 GB of GPU memory in bf16. Thinking is off. Trained with soft labels on 473k questions, about 223k of them from the train splits of benchmark families in the Decision Index; the maker says its index score is therefore largely in-distribution. Mogan reports Decision Index 0.2.1 at 65.42 from its own full run with the official kit; that figure is not on the third-party public board. Weakest on GPQA-Diamond, MMLU-Pro and HLE by the maker's account. English-centric. The temperature was fitted on held-out training data, and no calibration error is reported. Teacher labels came from a model under the Qwen Community License. Weights CC BY-NC 4.0 (non-commercial training data), so they are not mirrored or packaged here.
What it decides
- choice — picks one option from a set
- noul — answers a yes/no question with one probability
At a glance
| Parameters | 31B |
| Base model | google/gemma-4-31B-it |
| Maker | Mogan AI |
| Released | 2026-09-27 |
| License | cc-by-nc-4.0 |
| Reported latency | 107 ms median in-process per request on one GH200, one request at a time |
Get the weights
pip install systemonemodels
systemone pull moganai/mogan-decision
The files are served from the maker's Hugging Face repository, moganai/mogan-decision-31B-it, and verified against the checksums recorded here.
Read more
This page was opened by System One for Mogan AI, who can claim the organisation and take it over at any time.