lev (Interfaze AI, Qwen3.5-4B + LoRA r32, label-token readout + candidate-path head, per-bucket temperatures)
Newinterfaze-ai/lev
Compare the best decision models or read our decision model vs LLM guide.
This model is a typed decision model that returns calibrated probabilities for yes/no, multiple-choice, or score questions. It is based on the Qwen3.5-4B architecture and uses a LoRA rank of 32. It processes text inputs to produce decision outputs.
- decision model
At a glance
Specifications
What the API gives you
- Context window
- — tokens
- Max output
- —
- Input price
- — / 1M tokens
- Output price
- — / 1M tokens
- Cache read
- —
- Modalities
- text → decision
- Knowledge cutoff
- —
- Released
- —
Measured evidence
Public benchmark record
- JevBench
- 42.2 ± 5.62026-10-16
LMArena data licensed CC-BY-4.0. Aider (Apache-2.0) and SWE-bench (MIT) are open. Design Arena uses blind human preference.
Open-weight model on Hugging Face.
Task evidence
How broad is the support?
Each status is exact to this model version and the selected task. Official release claims remain provisional.
- Best Open-Weight Decision Model for Self-HostingProvisional evidence
100% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing. 1 of 1 configured evaluation feeds currently have results.
- Best Decision Model for Agent Routing and ScoringProvisional evidence
100% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing. 1 of 1 configured evaluation feeds currently have results.
Recommendation coverage
Where this model appears
Not currently included in a published recommendation. The task evidence panel above says exactly what is confirmed and what is still missing.
Catalog data via OpenRouter and models.dev. Last checked within the past 6 hours.