Typed decisions from Google's frozen Gemma-4-12B-it, with no fine-tuning. A small shim over unmodified vLLM letters the options, reads the model's own probability for each letter at one answer position and applies one calibration temperature; one output token per decision.
A Situated World Model for fast judgment in agent loops.
Decides
choice, score, noul
choice, route, classify
Architecture
cygnet
trio-spark
Fine-tuned from
google/gemma-4-12b-it
—
License
mit
proprietary
Availability
Open weights
Hosted API
Hosted by
—
MachineFi
Input price
—
$0.042/MTok
Decision accuracy
87.9%
—
Calibration error
—
—
Valid action rate
—
—
Median latency
50 ms
—
p95 latency
—
—
Evaluation suite
JevBench public set (231 items), JevBench CLI
—
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 cygnet and trio-spark-v1?
cygnet is from Blockbrain Labs and trio-spark-v1 from MachineFi. cygnet has open weights you can download and run; trio-spark-v1 is only available as a hosted API. Both answer choice questions. Only cygnet answers score and noul. Only trio-spark-v1 answers route and classify. cygnet reads up to 16K tokens of state, against 1K tokens for trio-spark-v1. cygnet is licensed mit; trio-spark-v1, proprietary.
Which is more accurate, cygnet or trio-spark-v1?
Only cygnet publishes an accuracy figure (87.9% on JevBench public set (231 items), JevBench CLI); trio-spark-v1 does not, so there is no comparison to make without your own test.
Which is cheaper, cygnet or trio-spark-v1?
cygnet: Free (open weights). trio-spark-v1: $0.042 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run cygnet or trio-spark-v1 locally?
cygnet yes — systemone pull blockbrain-labs/cygnet downloads its weights. The other is only served as a hosted API.
