DeepSeek V4 Preview Is Out — And It’s Picking Fights With OpenAI and Google

DeepSeek just released a preview of V4, their next big open-source model, and they’re not being shy about it. According to the company, V4 can go toe-to-toe with the best closed-source systems from Anthropic, Google, and OpenAI. That’s a bold claim, but given how much noise DeepSeek made last year, people are paying attention.

The headline improvement is in coding. DeepSeek says V4 is significantly better than earlier versions at writing and debugging code, which is exactly the kind of capability that powers tools like ChatGPT Codex and <a href="https://code.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code. If you’ve been watching the AI agent space, you know coding is the battleground right now — every major lab is racing to make models that can actually ship software without constant hand-holding.

What’s more interesting to me is the hardware angle. DeepSeek explicitly called out compatibility with Huawei’s domestic chips. That’s not just a technical footnote — it’s a political statement. China’s semiconductor industry has been under massive pressure from US export controls, and having a leading AI model run smoothly on homegrown hardware is a big flex. It suggests DeepSeek isn’t just competing on software; they’re helping validate China’s chip ecosystem.

This release comes roughly a year after DeepSeek first rattled US rivals. Back then, they showed that a Chinese company could build world-class models with less compute and more efficiency. V4 seems to double down on that philosophy — open-source, hardware-agnostic (within reason), and aggressively priced. I wouldn’t be surprised if Anthropic and Google are already stress-testing their own models against V4 benchmarks.

Of course, preview means it’s not final. We’ll have to wait for the full release and independent benchmarks to see if the claims hold up. But DeepSeek has earned the benefit of the doubt. They’ve consistently punched above their weight, and this feels like another step toward making open-source models a genuine threat to the closed-source giants.

If you’re building AI tools that need strong code generation, keep an eye on V4. And if you’re betting on China’s chip industry, this is as good a sign as any that the gap is closing.