DeepSeek debuts DeepSeek-V4.1-Flash, its smallest model built on a new Causal Encoder-Decoder architecture, with 552B backbone parameters and 1M-token context (Reuters)
DeepSeek is pitching V4.1-Flash as a cheaper, faster open-weight model for long-context agent work.
The model uses a 552B-parameter MoE backbone, with 8B active parameters for input and 16B for output. DeepSeek says the new Causal Encoder-Decoder design cuts KV cache needs to one-quarter the HBM and one-eighth the SSD storage of the prior generation. It also adds native visual understanding and supports up to a 1M-token context. Techmeme's note
The model uses a 552B-parameter MoE backbone, with 8B active parameters for input and 16B for output. DeepSeek says the new Causal Encoder-Decoder design cuts KV cache needs to one-quarter the HBM and one-eighth the SSD storage of the prior generation. It also adds native visual understanding and supports up to a 1M-token context. Techmeme's note
score 8