Media & Culture

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

DeepSeek V4 claims to rival GPT-5 and Claude on coding tasks.

Deep Dive

Chinese AI company DeepSeek released a preview of its next-generation model V4 on Friday, asserting it can compete with leading closed-source systems from US rivals such as Anthropic, Google, and OpenAI. The company emphasizes V4's significant improvements in coding—a capability central to AI agents that has driven tools like ChatGPT Codex and Claude Code. This open-source release marks a milestone for China's semiconductor industry, as DeepSeek explicitly highlights compatibility with domestic Huawei technology.

The preview arrives a year after DeepSeek's R1 model jolted US AI firms by claiming training at a fraction of the cost of leading systems. While DeepSeek has not disclosed V4's training costs or hardware, US officials have accused the company of using banned Nvidia chips, and Anthropic claims DeepSeek misused Claude to enhance its products. The release underscores escalating competition in AI development despite geopolitical tensions.

Key Points
  • DeepSeek V4 preview claims parity with US models from Google, OpenAI, and Anthropic.
  • Major coding improvements target AI agent capabilities like ChatGPT Codex and Claude Code.
  • Model supports domestic Huawei chips, a milestone for China's chip industry amid US export controls.

Why It Matters

DeepSeek's V4 challenges US AI dominance with open-source innovation, intensifying global competition.