DeepSeek-V4.1-Flash
Open in the interactive archive →
Overview
Efficiency-tier update to DeepSeek's V4 line: 552B MoE / 8B active, native vision, 1M context, 384K max output, 4x KV-cache reduction vs V4-Flash. MIT-licensed weights.
Generates
App available
Details
- Vendor
- DeepSeek
- Type
- Multimodal LLM
- Release year
- 2026
- Jurisdiction
- China
- CEO
- Liang Wenfeng
- Key staffers
- Liang Wenfeng (Founder & CEO, also co-founder of High-Flyer hedge fund)
- Deployment
- Cloud + Local (open weights, MIT license)
- Open source
- Yes
- Free tier
- Yes
- Subscription
- API: $0.30/M input, $1.20/M output (peak; ~32% input cut, 57% cache-read cut).
- Stock status
- Private (funded by High-Flyer hedge fund)
- X
- @deepseek_ai
Known safety issues
China-jurisdiction content alignment; see official model card.
AIDb Rating
AIDb v2.0 (curated): Capability 8/10 · Currency 10/10 · Access 10/10 · Significance 7/10 → 8.75
Last verified 2026-09-14 · Primary source ↗
Cite this
“DeepSeek-V4.1-Flash.” AIDb — The AI Database. Verified 2026-09-14. https://theaidatabase.com/m/deepseek-v4-1-flash.html
More from DeepSeek
DeepSeek Chat · DeepSeek Coder V2 · DeepSeek-Coder · DeepSeek-Coder-V2-Lite · DeepSeek-GRM-27B · DeepSeek-LLM · DeepSeek-R1 · DeepSeek-R1-0528