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.
Invergent's decision model for text and images. A fine-tune of Gemma-4-26B-A4B (about 4B parameters active per token) that answers Choice, Noul and Score questions, with an optional thinking mode for harder questions.
Decides
choice, score, noul, classify
choice, score, noul, classify, route
Architecture
jevembed
rune
Fine-tuned from
qwen/qwen3-embedding-4b
google/gemma-4-26b-a4b-it
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights + hosted API
Hosted by
—
Invergent
Input price
—
—
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 rune?
jevembed is from HIT-TMG (Lychee Team) and rune from Surogate (Invergent). jevembed has open weights you can download and run; rune has open weights and a hosted API. Both answer choice, score, noul and classify questions. Only rune answers route. rune reads up to 262K tokens of state, against 1K tokens for jevembed. jevembed is the smaller model, at 4.0B parameters to 26B.
Which is more accurate, jevembed or rune?
Only jevembed publishes an accuracy figure (85.9% on JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging); rune does not, so there is no comparison to make without your own test.
Which is cheaper, jevembed or rune?
jevembed: Free (open weights). rune: Hosted, price not published, or free to self-host. Open weights cost nothing per call beyond your own hardware.
Can I run jevembed or rune locally?
Yes, both: systemone pull hit-tmg/jevembed and systemone pull surogate/rune download the weights.
Evaluation suite
JevEmbed-Data test split (64,110 hard-label questions), final LoRA checkpoint before merging
