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.
meraGPT's hosted decision model (state-decider-1). Answers yes/no, choice and rubric-score questions over one state as calibrated distributions in a single pass, on the System One schema, so the typesafe-sdk works by changing its base URL.
Decides
choice, score, noul
choice, score, noul, classify, route
Architecture
djev
state-decider
Fine-tuned from
google/diffusiongemma-26b-a4b-it
—
License
apache-2.0
proprietary
Availability
Open weights + hosted API
Hosted API
Hosted by
Maisa
meraGPT
Input price
$0.035/MTok
$0.030/MTok
Decision accuracy
—
76.8%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
—
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 state-decider-1?
djev is from Maisa and state-decider-1 from meraGPT. djev has open weights and a hosted API; state-decider-1 is only available as a hosted API. Both answer choice, score and noul questions. Only state-decider-1 answers classify and route. djev is licensed apache-2.0; state-decider-1, proprietary.
Which is more accurate, djev or state-decider-1?
Only state-decider-1 publishes an accuracy figure (76.8% on typed-decisions benchmark (400 cases, 2,000 decisions; teacher-ensemble references), single run); djev does not, so there is no comparison to make without your own test.
Which is cheaper, djev or state-decider-1?
djev: $0.035 / $0 per 1M, or free to self-host. state-decider-1: $0.03 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run djev or state-decider-1 locally?
djev yes — systemone pull maisa/djev downloads its weights. The other is only served as a hosted API.
Evaluation suite
—
typed-decisions benchmark (400 cases, 2,000 decisions; teacher-ensemble references), single run
