Native MLX FP16 conversion of Laya for Apple silicon. The same ModernBERT-large encoder, decision transformer and heads, running in MLX with no PyTorch; the checkpoint Laya Studio fine-tunes from.
An open System One model from Interfaze AI. A rank-32 LoRA on Qwen3.5-4B reads yes/no, choice and score answers from the logits of one forward pass, returns calibrated probabilities and serves the /v1/systemone API.
Decides
choice, score, noul
choice, score, noul, classify, route
Architecture
laya
lev
Fine-tuned from
convai-innovations/laya
qwen/qwen3.5-4b
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
—
68.9%
Calibration error
—
0.115
Valid action rate
—
—
Median latency
—
69 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 laya-mlx and lev?
laya-mlx is from aac6fef and lev from Interfaze AI. Both have open weights you can download and run. Both answer choice, score and noul questions. Only lev answers classify and route. laya-mlx is the smaller model, at 421M parameters to 4.0B.
Which is more accurate, laya-mlx or lev?
Only lev publishes an accuracy figure (68.9% on S1Bench (13 public subsets, macro)); laya-mlx does not, so there is no comparison to make without your own test.
Which is cheaper, laya-mlx or lev?
laya-mlx: Free (open weights). lev: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run laya-mlx or lev locally?
Yes, both: systemone pull aac6fef/laya-mlx and systemone pull interfaze-ai/lev download the weights.
