DeepSeek-V4.1-Flash redesigns the Transformer for agents, cutting KV cache to 890 bytes/token
AndLukyane · x · 2026-09-21
- What it is: DeepSeek-V4.1-Flash rethinks the Transformer around agent workloads — coding, research, and tool-using agents process hundreds of thousands of tokens while emitting little output, making prefill compute and KV cache the key infrastructure bottleneck.
- Architecture: A 552B-parameter MoE backbone with a new Causal Encoder-Decoder (CED) design, splitting 40 layers into a 20-layer causal encoder and 20-layer decoder; only 8B parameters activate per token during prefill and 16B during decoding.
- Cache compression: Compressed Sparse Attention 2 (CSA2) shares KV representations and sparse-attention decisions across layers; SWA Bounded Replay avoids persistently storing sliding-window KV states; FP4 KV quantization shrinks the global cache to just 890 bytes per token — about a quarter of DeepSeek-V4-Flash's.
- Capabilities retained: 1M-token context, native vision, and strong agentic behavior. Trained on 45T multimodal tokens, with much of the gain coming from scaling auto-generated agent tasks, environments, and rollouts.
Related event: DeepSeek Unveils V4.1-Flash, a 552B MoE Redesigned for Agents(3 posts)→
More from Infra
- Cloudflare Talk: Swapping Some Agent LLM Calls for Fast Models Like Jev Gives Big Speedups — michellechen · 2026-11-12
- Cloudflare Trains Its Own Clef and Clairic Models to Complete Its Agent Stack — michellechen · 2026-11-12
- AWS open-sources 23 production-ready GenAI assets, cutting build time from months to weeks — AWS ML Blog · 2026-11-12
- Sixth Street CEO: ~$200B of Chips Sitting in Warehouses as Data Centers Lag Behind — SumitGup · 2026-11-12
- AWS CEO Matt Garman: Space Data Centers Are 10-20 Years Away, Likely Not Economically Feasible — rohanpaul_ai · 2026-11-12
- CoreWeave Database Goes GA: Managed PostgreSQL Running Inside Your CKS Cluster Next to GPUs — _ScottCondron · 2026-11-12