Hi, 5ak3t here, I built SCM. Thanks for stopping by. What it is: SCM (Sovereign Classification Model) is a family of open models that do one job: you give them a text and a question, and they return a probability for every answer. Choose from your options, get yes/no, or place the text on a scale. Why I built it: the best typed-decision models ask you to accept something. The hosted reference is accurate, but your text goes to someone else's servers and every call is billed. The strongest open ones need 27B-class hardware. The tiny fast ones give up accuracy and some quietly train on the benchmark's own data families. I wanted the middle: accurate enough to test, small enough to self-host, open enough to audit. What "open" means here: the weights, every training source with its licence, the sources we excluded and why, the prompt format and scoring recipe, and an audit of how much a public benchmark overlaps our training data, all published next to the results. The numbers, plainly: on S1MB English the 14B model completes all 137 benchmarks with zero failed decisions and scores 55.5 rank 19 of 165 by our computation. That's below the hosted Jev 1.13 reference (59.6) and below the 27B class. If you want an open, self-hostable model whose data and evaluation you can check, that's what SCM is for. Models: https://huggingface.co/5ak3t/scm... · https://huggingface.co/5ak3t/scm... Questions, scepticism and benchmark gripes all welcome, I'll be here all day.

SCM

Open models that turn text + a question into probabilities

Open models that turn text + a question into probabilities