DeepSeek released its V4.1 Flash model on September 10, 2026, introducing a new Causal Encoder-Decoder MoE architecture with 552B total parameters (8B active on input, 16B on output), native multimodal vision, 1M-token context, and MIT-licensed open weights. The update delivered strong gains on agentic benchmarks like Terminal-Bench 2.1 (90.6) and DeepSWE while undercutting prior V4 Pro pricing and performance on cost and speed. The company has signaled plans for a V4.1 Pro follow-on but has not announced timelines or previews since the September launch. With DeepSeek iterating roughly every three to four weeks historically, traders focus on credible leaks, API changelog updates, or developer conference signals that could indicate the next Flash variant. No major releases or official statements have emerged in the past month.