A Swiss and South African lab released hemmingway-1, a 27b open-weights model optimized for creative writing, roleplay, and dialogue, matching Claude fable 5's performance at 40x lower cost. It benchmarks at 1330 on eq-bench 4, outperforming GPT-5.5 and rivaling frontier models. The model targets users running local hardware and emphasizes frontier-level text generation capabilities.