5B uncensored reasoning model by WithIn Us AI. Abliterated with norm-preserving biprojection. 0.4% refusal rate.