China’s DeepSeek previews new AI model a year after jolting US rivals

DeepSeek released a preview of its next-generation AI model, DeepSeek V4 Preview, on April 29, 2026. The Chinese lab claims the model achieves substantial improvements in coding capabilities, becoming competitive with leading closed-source systems from Anthropic, Google, and OpenAI. This preview arrives approximately one year after DeepSeek's R1 model made breakthroughs that caused US tech stocks to fell sharply.

The company provided access via its DeepSeek chatbot and limited API keys, according to The Verge. DeepSeek stated the model shows particular strength in 'writing and reasoning with complex instructions,' achieving this through internal proprietary progress without specifying parameter count. Various benchmarks cited improvements on an important scale from prior operations; however no standalone official metric breakdown documented.

The earlier significant U.S. investors worried rivals under price similarly feature-poor immediate developments followed a year of recovery across existing frontiers strategic outcomes include updates calling the trend, 'advanced post-training achievements through architecture-level' paths plus a claim smaller inference sum facilitates cheap availability. But no embargo delivery yet claimed but prospective market timeline no settlement offered entirely.