An embedding model turned decision model. Qwen3-Embedding-4B with a merged LoRA, driven by the open JevEmbed toolkit, which scores Choice, Score and Noul questions from embedding similarity and returns a distribution without generating text.
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.
Decides
choice, score, noul, classify
choice, score, noul
Architecture
jevembed
lumma-fev
Fine-tuned from
qwen/qwen3-embedding-4b
frontiersmind/lumma-0.6b-base
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
85.9%
64.0%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
45.8 ms
p95 latency
—
—
Figures are from each model’s manifest; accuracy and latency are what the publishers report, on their own suites and hardware. Add a third model.
Questions
What is the difference between jevembed and lumma-fev?
jevembed is from HIT-TMG (Lychee Team) and lumma-fev from FrontiersMind. Both have open weights you can download and run. Both answer choice, score and noul questions. Only jevembed answers classify. lumma-fev reads up to 8K tokens of state, against 1K tokens for jevembed. lumma-fev is the smaller model, at 649M parameters to 4.0B.
Which is more accurate, jevembed or lumma-fev?
They report on different suites — jevembed 85.9% on JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging, lumma-fev 64.0% on typed-decisions (maker's table; split not stated) — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, jevembed or lumma-fev?
jevembed: Free (open weights). lumma-fev: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run jevembed or lumma-fev locally?
Yes, both: systemone pull hit-tmg/jevembed and systemone pull frontiersmind/lumma-fev download the weights.
Evaluation suite
JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging
