Jev-Omni
A multimodal decision classifier over text, image, audio and video on Gemma 4 12B, fine-tuned on 30,000 questions. Returns a probability per option and generates nothing.
On its own DecisionBench Medium it scores 87.6% with a calibration error of 0.040, and 86.2% on the matched JevBench subset. Latency on a warm H200: 83 ms for about 2,000 text tokens, 26 ms per image, 31 ms audio, 504 ms video; it is best with 20 options or fewer. The card declares Apache-2.0 "following Gemma 4"; Gemma bases normally carry Google's Gemma terms, so check before redistributing. The unified checkpoint is 24 GB; FP32 backbone shards add about 50 GB. Community GGUF and MLX quantisations exist.
What it decides
- choice — picks one option from a set
- noul — answers a yes/no question with one calibrated probability
- score — places the input on an ordered scale
- classify — assigns a category from a fixed taxonomy
At a glance
| Parameters | 12B |
| Base model | google/gemma-4-12b-it |
| Maker | akhilaaa3 |
| Released | 2026-09-20 |
| License | apache-2.0 |
| Reported accuracy | 87.6% |
| Reported latency | 83 ms text, 26 ms image, 31 ms audio, 504 ms video on a warm H200 |
Get the weights
pip install systemonemodels
systemone pull akhilaaa3/jev-omni
The files are served from the maker's Hugging Face repository, akhilaaa3/Jev-Omni, and verified against the checksums recorded here.
Read more
This page was opened by System One for akhilaaa3, who can claim the organisation and take it over at any time.