A typed-decision layer for Google's DiffusionGemma, from David Villalón at Maisa AI. It compiles a request into a small answer canvas, runs one denoising read on patched vLLM and reads the probabilities of the allowed labels, for text, images and images offered as options.
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
choice, score, noul
Architecture
djev
lumma-fev
Fine-tuned from
google/diffusiongemma-26b-a4b-it
frontiersmind/lumma-0.6b-base
License
apache-2.0
apache-2.0
Availability
Open weights + hosted API
Open weights
Hosted by
Maisa
—
Input price
$0.035/MTok
—
Decision accuracy
—
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 djev and lumma-fev?
djev is from Maisa and lumma-fev from FrontiersMind. djev has open weights and a hosted API; lumma-fev has open weights you can download and run. Both answer choice, score and noul questions. lumma-fev is the smaller model, at 649M parameters to 26B.
Which is more accurate, djev or lumma-fev?
Only lumma-fev publishes an accuracy figure (64.0% on typed-decisions (maker's table; split not stated)); djev does not, so there is no comparison to make without your own test.
Which is cheaper, djev or lumma-fev?
djev: $0.035 / $0 per 1M, or free to self-host. lumma-fev: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run djev or lumma-fev locally?
Yes, both: systemone pull maisa/djev and systemone pull frontiersmind/lumma-fev download the weights.
