DeepSeek's rapid iteration on its V4 family, including the September 10 release of V4.1-Flash with native multimodal support and a new causal encoder-decoder MoE architecture, sets expectations for a V4.1-Pro or successor soon. The company maintains a roughly 63-day average release cadence across open-weight models emphasizing agentic coding, 1M-token context, and efficiency on Huawei Ascend chips. Recent benchmarks show strong performance in tool-use and terminal tasks, while partnerships for CUDA alternatives and a $1B annualized revenue run rate underscore scaling momentum ahead of a potential IPO. No firm date exists for the next Pro variant, though historical patterns point to a possible November window amid competitive pressure from other Chinese labs.