DeepSeek open-sources a preview of its V4 model family

DeepSeek/DeepSeek-V series

DeepSeek released open-weight previews of V4-Pro and V4-Flash. V4-Pro has 1.6 trillion parameters with 49 billion active, while V4-Flash offers 284 billion parameters with 13 billion active. Both support 1-million-token context as standard via token compression and sparse attention, with thinking and non-thinking modes, and compatibility with OpenAI and Anthropic APIs.

• V4-Pro: 1.6T total / 49B active parameters; V4-Flash: 284B total / 13B active parameters for faster, economical inference
• Both models support 1-million-token context as default with token-wise compression and DeepSeek Sparse Attention mechanisms
• Support OpenAI ChatCompletions and Anthropic APIs; optimised for agent integration with Claude Code and OpenClaw
• Both offer dual thinking and non-thinking modes; available via API on April 24, 2026