DS
DeepSeekDeepSeek-V4-Flash
V4 high-speed, low-cost edition — strong reasoning, very low call cost
Parameters284B total (13B active per inference)Context Window1M tokens (max output 384K)Knowledge CutoffNot publicly disclosed
Overview
DeepSeek-V4-Flash is the high-speed, low-cost edition in the V4 flagship line, pairing strong reasoning with very low invocation cost for high-frequency production use.
It complements V4-Pro within the latest V4 generation: Flash emphasizes throughput and economy where Pro emphasizes peak performance, both under the same low-cost positioning.
DeepSeek-V4-Flash reinforces that high-speed, low-cost tiers can still deliver strong reasoning performance, illustrating that relatively low-cost models can achieve high performance at scale.
Use Cases
Content WritingCode ProgrammingLogical Reasoning
CHINA LLM GUIDES
Chinese LLM selection guides
- DeepSeek V4 Guide 2026: V4-Pro vs V4-Flash vs R1DeepSeek-V4-Pro is the reasoning flagship on this index: 1.6T total parameters, about 49B active, 1M context, 384K max output. V4-Flash is the smaller, faster 284B / 13B-active twin with the same context. R1 is the separate chain-of-thought model.Read article →
- Chinese LLM Context Window 2026: Who Publishes 1M TokensOn this index, four pages publish a 1M-token context: DeepSeek-V4-Pro, DeepSeek-V4-Flash, Qwen-Long, and Qwen3.7-Max. Qwen3-Max, GLM-4.7, and Kimi K2.6 do not. Quote only the published figure.Read article →