A training-free layer that turns an open LLM into a decision model. It reads typed choice, yes/no and score answers from one prefill, removes option-order bias with no labels and, from a few hundred labels, calibrates or fits a closed-form head. Weights stay untouched.
Ultra-small English encoders (17M, 68M and 395M) fine-tuned from Ettin rerankers. Each answers choice, noul and score questions with a probability per candidate and no generated text. Experimental v0 by one developer; the weights have no licence assigned yet.
Decides
choice, score, noul, classify, route
choice, noul, score
Architecture
anyjev
bekko
Fine-tuned from
qwen/qwen3-8b
cross-encoder/ettin-reranker-400m-v1
License
apache-2.0
Unassigned: the weights licence is not finalised (treated as non-commercial); training code MIT
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
77.1%
—
Calibration error
0.034
—
Valid action rate
—
—
Median latency
—
5 ms
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 anyjev and bekko-system-one?
anyjev is from Nokia and bekko-system-one from Yuichi Tateno (hotchpotch). Both have open weights you can download and run. Both answer choice, score and noul questions. Only anyjev answers classify and route. bekko-system-one is the smaller model, at 395M parameters to 8.0B. anyjev is licensed apache-2.0; bekko-system-one, other.
Which is more accurate, anyjev or bekko-system-one?
Only anyjev publishes an accuracy figure (77.1% on LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question); bekko-system-one does not, so there is no comparison to make without your own test.
Which is cheaper, anyjev or bekko-system-one?
anyjev: Free (open weights). bekko-system-one: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run anyjev or bekko-system-one locally?
Yes, both: systemone pull nokia/anyjev and systemone pull hotchpotch/bekko-system-one download the weights.
—
5.4 ms
Evaluation suite
LocalLLaMA/typed-decisions (2,000 held-out decisions), L2 with 300 labels per question
The maker's short-input speed probe (4-option Choice, RTX 5090, 1 Oct 2026)
