DeepSeek releases V4.1 Flash: 552B MoE beats V4 Pro, cuts prices, retires flagship
DeepSeek · wechat · 2026-09-10
DeepSeek officially launched V4.1 Flash, the smallest model in its new architecture family: a 552B MoE with a novel asymmetric Causal-Encoder-Decoder design (8B input / 16B output activations) and native vision understanding.
- Benchmarks beat DeepSeek V4 Pro and other flagships; V4 Pro will be retired, with requests to deepseek-v4-pro routed to V4.1 Flash from Sept 14.
- KV cache shrunk 4x vs the previous generation (1/4 HBM, 1/8 SSD needs), 437x vs the first model, sharply cutting agent-serving costs.
- New pricing takes effect Sept 10 with off-peak rates at half price; API name is deepseek-flash.
- Tencent WorkBuddy, CodeBuddy and OpenCode are launch partners; the model is MIT-licensed with a tech report, and the team is seeking large-scale deployers (2k GPUs + storage cluster).
More from Infra
- Michael Dell: Musk is "the most demanding customer there is" — Colossus servers built in ~6 weeks — elonmusk · 2026-10-27
- AI chip startup Tenstorrent raises $1.5B at $6.2B valuation, more than doubling — Beth_Kindig · 2026-10-27
- Ex-MediaTek expert: Google TPU external revenue $8B this year, $104B by 2028 — RihardJarc · 2026-10-27
- Anthropic discloses up to $120.5B SpaceX compute deal, ~$3.1B monthly at full capacity — XFreeze · 2026-10-27
- Satire: swapping Raft for an LLM-style decision model tanks write throughput 99.98% — irombie · 2026-10-27
- Cadence joins the domain-specific AI model camp for chip design — karlfreund · 2026-10-27