TechFlow News: On April 24, DeepSeek officially launched and open-sourced the preview version of its new model series, DeepSeek-V4. The model comes in two variants: DeepSeek-V4-Pro, whose performance rivals top-tier closed-source models and leads the open-source field in agent capabilities, world knowledge, and mathematical/coding reasoning; and DeepSeek-V4-Flash, a more compact model offering faster, more cost-effective API services. Both models support a 1M (million)-token ultra-long context and operate in both non-thinking and thinking modes.
Technically, DeepSeek-V4 introduces a novel attention mechanism that combines DSA sparse attention—achieving globally leading long-context capability while significantly reducing computational and memory requirements. For APIs, the model parameter names are “deepseek-v4-pro” and “deepseek-v4-flash”; the legacy model names “deepseek-chat” and “deepseek-reasoner” will be deprecated on July 24, 2026.
