DeepSeek's V4.1 Flash, released September 10, 2026, introduced a new Causal Encoder-Decoder MoE architecture with 552B total parameters (8B-16B active), native multimodal vision, 1M-token context, and sharply reduced KV cache needs. It outperformed the prior V4 Pro on multiple agentic and coding benchmarks while cutting API costs, leading to the retirement of older Flash variants and routing of Pro traffic to the new model. As of mid-October 2026, no official announcements or credible reports have surfaced on the next Flash iteration or V4.1 Pro timeline. Traders are watching for DeepSeek's IPO preparations and any competitive responses from labs like Alibaba or OpenAI that could accelerate follow-on releases.