Lumma-Fev-0.6B
FrontiersMind's decision model on a base it pre-trained from scratch. Each question reads the state and its own tokens once and a 256-dimension pointer head scores the options; Choice and Score take up to 255 entries, and nothing is generated.
A merged rank-16 LoRA on Lumma-0.6B-Base, for English and ten Indian languages, with a pip package (lumma-fev) and a /v1/systemone server. The maker's table gives 0.64 on typed-decisions, 0.84 on AG News and 0.90 on Banking77 for this size, and 0.78 and 0.81 on typed-decisions for Lumma-Fev-4B and 9B (released 25 September; their cards do not name a base). The table does not say which split or hardware was used. The third-party Decision Index 0.2.1 gives the 0.6B a balanced score of 2.97, so treat the maker's figures as unconfirmed.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one calibrated probability
At a glance
| Parameters | 649M |
| Base model | frontiersmind/lumma-0.6b-base |
| Maker | FrontiersMind |
| Released | 2026-09-22 |
| License | apache-2.0 |
| Reported accuracy | 64.0% |
| Reported latency | 45.8 ms p50 (hardware not stated) |
Get the weights
pip install systemonemodels
systemone pull frontiersmind/lumma-fev
The files are served from the maker's Hugging Face repository, FrontiersMind/Lumma-fev-0.6b, and verified against the checksums recorded here.
Read more
This page was opened by System One for FrontiersMind, who can claim the organisation and take it over at any time.