ModelsSep 10, 2026 · 15:408 sources

DeepSeek Replaces Flagship V4-Pro With 552B Parameter V4.1-Flash Model

From huggingnews.com

Image: huggingnews.com
AI;DR

DeepSeek has replaced its V4-Pro flagship with the V4.1-Flash model, a 552B parameter Mixture-of-Experts system using a new Causal Encoder-Decoder architecture with 8B active input and 16B output parameters. The model supports native multimodal input and a 1 million token context window, and all V4-Pro requests will route to V4.1-Flash at Flash rates starting September 14, 2026. The system's KV cache requires 1/4 of the HBM and 1/8 of the SSD storage of its predecessor, and updated API pricing launched September 10, 2026, maintaining a 50% discount for off-peak workloads.

Read the sourceSign in to suggest

Sources

+ 3 more
  1. SUPPORT@vllm_project“Engram: a quarter of the checkpoint is n-gram memory the model looks up instead of computes”x.com
  2. SUPPORT@eliebakouch“trained on 45T tokens”x.com
  3. SUPPORT@reuters“China's DeepSeek launches V4.1-Flash model”x.com
#deepseek#llm#multimodal#inference#benchmark#open-source#v41-flash#chipsMarkdown
Content history (2)
  • 2026-10-04 · Summary · vi · Replaced stored text
    DeepSeek vừa ra mắt mô hình V4.1-Flash với 552 tỷ tham số, thay thế phiên bản V4-Pro trướ…
    Kiến trúc Causal Encoder-Decoder mới sử dụng 8B tham số hoạt động cho đầu vào và 16B cho …
  • 2026-10-04 · Title · vi · Replaced stored text
    DeepSeek thay thế V4-Pro bằng mô hình V4.1-Flash 552 tỷ tham số
    DeepSeek thay thế V4-Pro chủ lực bằng mô hình V4.1-Flash 552B tham số

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today