DeepSeek released V4.1 Flash on September 10, 2026, as its current Flash-tier model—a 552B MoE with new Causal Encoder-Decoder architecture, 8B/16B active parameters, native multimodal vision, 1M-token context, and MIT-licensed weights. It retired the prior V4 Flash, improved agentic benchmarks like Terminal-Bench 2.1 to 90.6, and undercut its own V4 Pro on cost and speed while maintaining strong performance against closed models. DeepSeek has iterated rapidly on the Flash line, with releases roughly every 20–60 days in 2026 amid broader monthly updates. No official timeline or announcement exists for the next Flash variant or V4.1-Pro as of mid-October, though historical cadence and ongoing efficiency focus (KV cache compression, multimodal scaling) suggest continued fast iteration. Traders weigh this pattern against typical development timelines and competitive pressure from other labs.