Together AI's Jev-style classifier. A LoRA fine-tune of Qwen3.5-4B that reads a state, a question and 2 to 24 options and returns one option letter. Served on Together's platform; recipe published.
A ModernBERT-large encoder with an option-marker head. Premise and options are packed into one sequence and each option's marker is scored in a single bidirectional pass; version 1.2 makes the scoring order-invariant.
Decides
choice, classify, route
choice, score, noul, classify
Architecture
tev
von
Fine-tuned from
qwen/qwen3.5-4b
answerdotai/modernbert-large
License
Unspecified — weights licence being finalised
apache-2.0
Availability
Open weights + hosted API
Open weights
Hosted by
Together AI
—
Input price
$0.042/MTok
—
Decision accuracy
88.0%
63.9%
Calibration error
—
0.045
Valid action rate
—
—
Median latency
—
18 ms
p95 latency
—
—
Evaluation suite
Together development set (reused, not held out)
JevBench public standard tier
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 tev1 and von?
tev1 is from Together AI and von from wfzyx. tev1 has open weights and a hosted API; von has open weights you can download and run. Both answer choice and classify questions. Only tev1 answers route. Only von answers score and noul. von is the smaller model, at 395M parameters to 4.0B. tev1 is licensed other; von, apache-2.0.
Which is more accurate, tev1 or von?
They report on different suites — tev1 88.0% on Together development set (reused, not held out), von 63.9% on JevBench public standard tier — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, tev1 or von?
tev1: $0.042 / $0 per 1M, or free to self-host. von: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run tev1 or von locally?
Yes, both: systemone pull together-ai/tev1 and systemone pull wfzyx/von download the weights.
