DeepSeek V3.1 Terminus

V3.1-Terminus is built to prioritize correctness and step-by-step logic over speed. It absorbs large prompts—multi-document briefs, repositories, and long chats—then plans, checks intermediate steps, and integrates retrieved evidence before composing an answer. The model follows instructions precisely, emits schema-true JSON when automations require it, and calls tools to run calculations, query knowledge bases, or execute code in the loop. Bilingual by design, it handles Chinese and English comfortably across technical and business domains. In practice, teams route routine queries to lighter siblings and reserve Terminus for the hardest tasks: complex analytics, formal reasoning, repository-scale refactors, policy-aware drafting, and investigations where grounded, auditable outputs matter more than latency.