Valen (万澜) is a multimodal decision model from Liuziyu77 that brings visual perception to System One decision-making. Inspired by TypeSafe's Jev, it evaluates text, images and video against task instructions and returns probabilities over supplied candidates, giving software a structured decision interface. A Qwen3.5-0.8B or 2B backbone and a shared decision head score candidates without generating answer tokens. The repository includes the model implementation, data processing, SFT and experimental RLCD training, and inference and evaluation commands, with support for training on your own data. The released Valen-Preview-0923 checkpoint is a 2B Sokoban-trained preview built on Qwen3.5-2B: it starts from a General 100k SFT decision head, then runs RLCD on 30,000 Sokoban records for three epochs. Its model card reports 87.60% accuracy on 500 single-step Sokoban action questions and 38 of 100 complete games solved (at most 200 moves), on a selected set of 35 easy and 65 medium levels that is conditioned on model outcomes rather than an unbiased benchmark. In the README demo, the preview solves one Sokoban level in 9 decisions with 1.13 seconds of cumulative decision latency, while Qwen3.8-27B-FP8 takes 198.05 seconds in thinking mode and fails in no-thinking mode. The README's General-versus-Sokoban comparison figure covers Qwen3.5-0.8B, Qwen3.5-2B and Valen 2B RL checkpoints, with the General panel using a separate General-trained RL checkpoint rather than the Sokoban preview. Running the preview requires both the Valen checkpoint and the Qwen3.5-2B base model. Code is Apache-2.0; base models and source datasets keep their own licenses.