Announced a week ago, DeepSeek-R1 — a model which is said to use more concise “reasoning patterns”, less data, and lower-cost ...