REDDIT · REDDIT// 3h agoMODEL RELEASE

DeepSeek V4 Flash undercuts Gemini 3

DeepSeek released V4 Flash, an open-weights 284B parameter MoE model offering near-frontier reasoning and a 1 million token context window. At just $0.14 per million input tokens, it drastically reduces the cost of high-speed agentic tasks.

// ANALYSIS

DeepSeek V4 Flash forces a new pricing bottom for fast, capable inference models.

–Priced at $0.14/$0.28 per million tokens, it aggressively undercuts GPT-5.4 Nano and Claude Haiku 4.5
–The massive 1 million token context window, enabled by Compressed Sparse Attention, makes large-scale document processing economically viable
–As an MIT-licensed open-weights release, it accelerates local development and pressures closed-source API pricing

// TAGS

deepseek-v4llmagentopen-weightspricing

DISCOVERED

3h ago

2026-04-25

PUBLISHED

6h ago

2026-04-25

RELEVANCE

10/ 10

AUTHOR

NoFaithlessness951