TLDR Tech / Launches
DeepSeek's V4.1 Flash 1017: A sparse MoE model with image capability
DeepSeek has launched V4.1 Flash 1017, its initial sparse mixture-of-experts model utilizing its Causal Encoder-Decoder framework. This model engages 8B parameters for input and 16B for output from a 552B-parameter base, incorporating native image processing and optimized KV caching for economical agent and extended context tasks.
Read the full story