DeepSeek-V4 is engineered to handle ultra-long contexts, capable of processing up to one million words. It leads the pack in both the domestic and open-source arenas, excelling in Agent capabilities, world knowledge, and reasoning performance. The API service for this model has been concurrently updated, enabling users to effortlessly integrate it by simply adjusting the model_name parameter. In the Agentic Coding assessment, the V4-Pro variant showcased outstanding performance, reaching the pinnacle among currently available open-source models. It also performed exceptionally well in other Agent-related evaluations. Presently, DeepSeek-V4 has emerged as the top choice for Agentic Coding among company employees, offering a markedly superior user experience when compared to Sonnet 4.5. Its performance quality closely rivals that of Opus 4.6's non-thinking mode, although there remains a notable difference when pitted against Opus 4.6's thinking mode.