Skip to content
View original post on X: ModelScopeOfficial· 38/100AI score38/100

JEV-35B, a fast decision model built on Qwen3.5-35B-A3B, released under Apache 2.0

AISummary

ModelScope releases JEV-35B, a System 1 decision model built on Qwen3.5-35B-A3B that activates about 3B of its 35B parameters per token, under Apache 2.0.

It compares up to 256 text or image options in one pass, returns calibrated probabilities, and reaches about 204ms median single-request latency in its tested setup.

The NVFP4 version is 25.6GB and runs on a single 32–48GB GPU, preserving 96.6% of the BF16 model's decisions.

Post on XView on X
ModelScopeVerified on X
@ModelScope2022

JEV-35B turns Qwen3.5-35B-A3B into a fast System 1 decision model, activating only around 3B of its 35B parameters per token. Apache 2.0.
🤖 https://modelscope.ai/collections/autotrust/JEV-35B

🏆 Outperforms JEV-27B-VL on the Decision Index public suite while retaining competitive visual decision performance.
⚡ Compare up to 256 text or image options in one pass, return calibrated probabilities, and reach around 204ms median single-request latency in the tested setup.
📦 The NVFP4 version is only 25.6GB and uses around 23GiB for weights, enabling local deployment on a single 32–48GB GPU.
🔍 Quantization preserves 96.6% of the BF16 model’s decisions with nearly unchanged latency and only a small benchmark drop.
🛠️ Start with the included vLLM script. Use /v1/decide for fast System 1 decisions or OpenAI-compatible endpoints for System 2 reasoning.

Source: ModelScope · x.comPublished

Topics

Open-source ecosystemModel releasesEvals & benchmarks