DeepSeek V4 nears April launch

If the leaks hold, V4 looks less like a routine model bump and more like DeepSeek doubling down on the two things developers care about most: code quality and context depth. The Huawei-first infrastructure angle is the bigger strategic tell, because it suggests DeepSeek is building for domestic deployment realities, not just benchmark bragging rights.

• –Dataconomy cites Whale Lab reports that V4 is a multimodal model for text, image, and video generation
• –The article says coding and long-term memory are core upgrades, which would make V4 a direct competitor to the current frontier coding models
• –DeepSeek’s homepage still promotes V3.2, so V4 remains unannounced and this should be treated as reported, not confirmed
• –A Huawei/Cambricon optimization path could matter more than raw model size for enterprise adoption in China