DeepSeek Reworks Frontier Model Economics With V4.1 Flash

DeepSeek V4.1 Flash reportedly combines 552B parameters with asymmetric activation, compressed KV caching, recomputation, a one-million-token context, and MIT-licensed open weights to reduce inference costs for long-context and agent workloads. The launch and retirement of V4 Pro could materially affect open-model deployment economics, but independent benchmarks, production throughput, and scaffold sensitivity remain unresolved.