An embedding model turned decision model. Qwen3-Embedding-4B with a merged LoRA, driven by the open JevEmbed toolkit, which scores Choice, Score and Noul questions from embedding similarity and returns a distribution without generating text.
Upstage's structured decision model, a System One endpoint on Solar Mini 4. Returns a choice, a score or a yes/no answer with a probability read from the model, in one forward pass, on the same /v1/systemone schema as Jev.
Decides
choice, score, noul, classify
choice, score, noul, classify, route
Architecture
jevembed
solar
Fine-tuned from
qwen/qwen3-embedding-4b
—
License
apache-2.0
proprietary
Availability
Open weights
Hosted API
Hosted by
—
Upstage
Input price
—
$0.10/MTok
Decision accuracy
85.9%
—
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 jevembed and solar-decide?
jevembed is from HIT-TMG (Lychee Team) and solar-decide from Upstage. jevembed has open weights you can download and run; solar-decide is only available as a hosted API. Both answer choice, score, noul and classify questions. Only solar-decide answers route. solar-decide reads up to 512K tokens of state, against 1K tokens for jevembed. jevembed is licensed apache-2.0; solar-decide, proprietary.
Which is more accurate, jevembed or solar-decide?
Only jevembed publishes an accuracy figure (85.9% on JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging); solar-decide does not, so there is no comparison to make without your own test.
Which is cheaper, jevembed or solar-decide?
jevembed: Free (open weights). solar-decide: $0.1 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run jevembed or solar-decide locally?
jevembed yes — systemone pull hit-tmg/jevembed downloads its weights. The other is only served as a hosted API.
Evaluation suite
JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging
