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.
FLock.io's typed-decision model, fine-tuned from Mapika's decider-2b. Reads one hidden state per question and scores the declared options, so an answer outside the list cannot occur; several questions about one state share a forward pass.
Decides
choice, score, noul
choice, score, noul
Architecture
djev
this-that
Fine-tuned from
google/diffusiongemma-26b-a4b-it
mapika/decider
License
apache-2.0
mit
Availability
Open weights + hosted API
Open weights
Hosted by
Maisa
—
Input price
$0.035/MTok
—
Decision accuracy
—
87.8%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
31.4 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 this-that-model?
djev is from Maisa and this-that-model from FLock.io. djev has open weights and a hosted API; this-that-model has open weights you can download and run. Both answer choice, score and noul questions. this-that-model is the smaller model, at 1.9B parameters to 26B. djev is licensed apache-2.0; this-that-model, mit.
Which is more accurate, djev or this-that-model?
Only this-that-model publishes an accuracy figure (87.8% on limberc/this-that-complex-decisions (1,710 questions, FLock's benchmark)); djev does not, so there is no comparison to make without your own test.
Which is cheaper, djev or this-that-model?
djev: $0.035 / $0 per 1M, or free to self-host. this-that-model: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run djev or this-that-model locally?
Yes, both: systemone pull maisa/djev and systemone pull flock-io/this-that-model download the weights.
