DeepSeek unveils V4 Pro and V4 Flash with 1M token context

DeepSeek has officially entered public preview with its next-generation AI models—V4 Pro and V4 Flash. V4 Pro packs a staggering 1.6 trillion parameters, while the more efficient V4 Flash offers 284 billion parameters, making both models significantly more capable in knowledge, reasoning, and multi-step task execution. A key highlight is the 1 million token context window, enabling processing of massive documents, codebases, or entire conversation histories without truncation. Both models also feature enhanced 'agentic' abilities, allowing them to autonomously plan and execute complex tasks—a critical upgrade for enterprise automation.

Developers can now access V4 Pro and V4 Flash through the DeepSeek API. DeepSeek has set a sunset date of July 24, 2026, for legacy API model names (deepseek-chat and deepseek-reasoner), pushing users to migrate to the new endpoints. This transition marks a major shift in DeepSeek's product lineup, positioning the company to compete directly with other frontier models like GPT-4o and Claude 3.5. With massive parameter counts and extended context, V4 Pro targets heavy-duty analytic workloads, while V4 Flash offers a cost-efficient alternative for high-throughput applications.