According to a report by the Financial Times on February 28, DeepSeek will release its new multimodal large language model V4 next week. The model natively supports image, video, and text generation. It is the first major update since the release of the R1 inference model in January this year, aiming to fill the market gap for high-performance, low-cost open-source multimodal models domestically.
According to the roadmap, DeepSeek will provide a brief technical explanation upon launch and release a detailed engineering report one month later. On the underlying ecosystem side, the V4 model has completed deep hardware adaptation optimization with Huawei and Cambricon. This move is expected to significantly boost domestic computing power demand and marks the acceleration of AI model inference processes toward domestic chip arrays.
As a strong competitor in the open-source community, the release of DeepSeek V4 indicates further lowering of the barriers for multimodal technology, and holds important implications for building an independent and controllable local AI computing ecosystem.
