All AI Models
Organization: DeepSeek

DeepSeek-V4.1-Flash

DeepSeek's V4.1-Flash-1017 iteration featuring an Asymmetric Causal MoE architecture that drastically shrinks KV-cache memory footprints for multi-file software engineering.

Official Release
Model TypeOpen-Weight Asymmetric Causal MoE
Context Window256,000 tokens
License / WeightsMIT License
API PricingMIT Open Weights · $0.14 / 1M input · $0.55 / 1M output
Reported Evaluation Benchmarks:
DeepSWE: High Efficiency
KV-Cache Footprint: Ultra-Compact Asymmetric MoE

Latest Releases, Benchmarks, Pricing & API News — DeepSeek-V4.1-Flash