A rank-16 LoRA on Qwen3.5-4B built for the JevBench setting. A document, a policy and a question go in; a calibrated distribution over the option letters comes out of one forward pass. Research and demo use only.
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.
Decides
choice, score, noul, classify
choice, score, noul
Architecture
hopper
laya
Fine-tuned from
qwen/qwen3.5-4b
convai-innovations/laya
License
Research and demo use only (training data includes RACE, non-commercial); serving code Apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
68.5%
—
Calibration error
0.102
—
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 hopper and laya-mlx?
hopper is from HopitAI and laya-mlx from aac6fef. Both have open weights you can download and run. Both answer choice, score and noul questions. Only hopper answers classify. laya-mlx is the smaller model, at 421M parameters to 4.0B. hopper is licensed other; laya-mlx, apache-2.0.
Which is more accurate, hopper or laya-mlx?
Only hopper publishes an accuracy figure (68.5% on JevBench public hard tier (111 items, measured by the authors)); laya-mlx does not, so there is no comparison to make without your own test.
Which is cheaper, hopper or laya-mlx?
hopper: Free (open weights). laya-mlx: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run hopper or laya-mlx locally?
Yes, both: systemone pull hopit-ai/hopper and systemone pull aac6fef/laya-mlx download the weights.
Evaluation suite
JevBench public hard tier (111 items, measured by the authors)
