DeepSeek V4 architecture reportedly redesigned: more stable and efficient, not simpler
teortaxesTex · x · 2026-10-21
DeepSeek watcher Teortaxes comments on reports that V4 uses a redesigned architecture said to be stable, scale more predictably, and be vastly more efficient — though not clearly simpler. Attached screenshots show community debate over choices like sqrt(softplus(·)), HashMoE, and compute budgets, with some calling V3.2's DSA the most elegant public LLM architecture. Unverified speculation ahead of V4.1 Pro.
More from Models
- Claude subscription margins questioned as $10 OpenCode Go plan swallows 3.4B tokens a month — teortaxesTex · 2026-10-21
- Dev slams Anthropic's auto mode classifier: stops zero bad actions, blocks legit work — cjimti · 2026-10-21
- bekko-static-embedding-v1: GPU-free static embeddings hit retrieval SOTA, 100x faster on CPU — tomaarsen · 2026-10-21
- Qwen3.8-Flash-Next runs stable at 1.66bpw exl3, fits entirely on one 5090 — Unstable_Llama · 2026-10-21
- User documents several real dangerous Gemini incidents in one thread — Regular_Manager_6235 · 2026-10-21
- MiniMax's M3.1 Flash talk: a faster model isn't faster if the job isn't — AI Engineer · 2026-10-21