
Featured
DeepSeek V4.1 Flash: China’s New AI Price-Performance Revolution
@kimmonismus presents DeepSeek V4.1 Flash as a rapid leap beyond July’s release, featuring a causal encoder-decoder architecture, a 552-billion-parameter MoE, and a far smaller KV cache. @deepseek_ai confirms the key figures, the phaseout of V4 Pro, API pricing, cache costs, and native multimodal support. While readers praise local inference and the release pace, mixed Terminal Bench results challenge its frontier-model status. Self-hosting is feasible for well-equipped deployments, though the official announcement also mentions 2,000-GPU installations with storage.




