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.
A Gemma 4 12B fine-tune for typed decisions that also chats and reads images. Ships as GGUF for llama.cpp, holds a 64K context on a 16 GB GPU, and serves /v1/systemone next to /v1/chat/completions.
Decides
choice, score, noul, classify, route
choice, score, noul, classify, route
Architecture
state-decider
winnow
Fine-tuned from
—
google/gemma-4-12b-it
License
proprietary
apache-2.0
Availability
Hosted API
Open weights
Hosted by
meraGPT
—
Input price
$0.030/MTok
—
Decision accuracy
76.8%
85.7%
Calibration error
—
—
Valid action rate
—
—
Median latency
—
143 ms
p95 latency
—
—
Evaluation suite
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 state-decider-1 and winnow?
state-decider-1 is from meraGPT and winnow from EldanRing. state-decider-1 is only available as a hosted API; winnow has open weights you can download and run. Both answer choice, score, noul, classify and route questions. winnow reads up to 64K tokens of state, against 4K tokens for state-decider-1. state-decider-1 is licensed proprietary; winnow, apache-2.0.
Which is more accurate, state-decider-1 or winnow?
They report on different suites — state-decider-1 76.8% on typed-decisions benchmark (400 cases, 2,000 decisions; teacher-ensemble references), single run, winnow 85.7% on JevBench public subset (231 items), Q8_0 — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, state-decider-1 or winnow?
state-decider-1: $0.03 / $0 per 1M. winnow: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run state-decider-1 or winnow locally?
winnow yes — systemone pull eldanring/winnow downloads its weights. The other is only served as a hosted API.
typed-decisions benchmark (400 cases, 2,000 decisions; teacher-ensemble references), single run
