Interfaze open-sources lev, a Qwen-3.5 4B scoring model
Probability-based scoring turns evaluation workloads into a batched classification task, cutting latency compared with having an LLM generate each judgement as text. Teams building automated grading, data filtering or LLM-as-judge pipelines can run it locally and avoid constrained-decoding overhead. Its foundation on Qwen-3.5 4B also makes it a lightweight alternative to jev for self-hosted setups.
Interfaze released lev, an open-source model described as an open version of jev.
It is built on Qwen-3.5 4B and answers true/false, multiple-choice and scoring questions about a given text without generating text token by token, instead reading results directly from the model's internal probabilities.
Multiple questions share a single forward pass, so total runtime stays close to a single query, and answer choices are strictly constrained to the given options.
Interfaze开源lev：开源版jev
基于Qwen-3.5 4B，能针对一段文本同时回答判断题、单选题和打分题，无需逐字生成内容，直接通过模型内部概率输出结果。多道题目共享一次计算，耗时和单次提问几乎一致，且选择范围被严格限定在给定的选项内。
模型：https://t.co/EGVp0Xuw6X
This report is categorized under AITOP's "Models" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include lev, Qwen-3.5, Interfaze, jev.
Source attribution: curated from Gorden Sun (@Gorden_Sun). Cross-source corroboration: 4 independent sources have covered this development (including AIGCLINK, OpenRouter, DAIR.AI), confirming industry-wide signal strength.
