Apps & data pipelines
FrontiersMind releases Lumma-Fev decision models and a Python client, with the 27B card documenting multilingual training from Qwen3.8 and fixed yes/no or choice answers.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The maker reports multilingual training and comparisons with other decision models; those results have not been independently reproduced. The 27B card documents yes/no and choice training, an 8,192-token context and substantial GPU memory needs. It is a separate Apache-2.0 family, not Jev weights.
Read the card for the exact model size you intend to run. The 27B release documents yes/no and choice training, but that does not establish every capability for every family member. Check the memory requirements and try your own known-answer cases before adopting its reported comparisons.