**DeepSeek released V4.1-Flash on September 10, 2026, as the first model in a new Causal Encoder-Decoder MoE architecture family (552B backbone parameters, activating ~8B on input and ~16B on output), adding native multimodal image understanding, a 1M-token context window, MIT-licensed open weights, and significantly lower API pricing than prior tiers.** This update retired the previous V4-Flash variants, outperformed the company's own V4-Pro on multiple agentic and coding benchmarks (e.g., Terminal-Bench 2.1 at 90.6, DeepSWE at 74.2), and led to a temporary reroute plan for Pro traffic that was later reversed due to user demand. DeepSeek has maintained a fast cadence in 2026, with V4 previews in April, a July Flash update, August Pro GA, and the September architecture refresh—roughly every 30–60 days recently—while preparing for a Shanghai STAR Market IPO. As of mid-October 2026, no confirmed announcements exist for the next Flash variant or V4.1-Pro, though analysts note the pattern and project potential follow-ups in the coming weeks. Key variables for timing include continued internal testing, competitive pressure from other Chinese labs on cost-efficient multimodal models, and any regulatory or financing milestones. Traders monitor DeepSeek's API changelog, Hugging Face releases, and official X posts for the earliest signals.