DeepSeek V4 Preview Is Here, and It’s Going After Claude and GPT Directly

DeepSeek just did it again. On Friday, the Chinese AI lab released a preview of V4, the long-awaited follow-up to the model that sent shockwaves through Silicon Valley a year ago. And this time, they’re not just playing catch-up—they’re aiming straight at the closed-source heavyweights.

The company claims V4 is a serious leap forward, especially in coding. That’s the battleground everyone cares about right now, because better code generation is what makes AI agents actually useful. Think ChatGPT Codex, <a href="https://chat.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code—DeepSeek wants a piece of that action, and they’re betting open-source can win.

What’s interesting here isn’t just the model itself. DeepSeek explicitly called out compatibility with Huawei’s hardware. That’s a big deal for China’s chip industry, which has been scrambling to reduce reliance on Western suppliers after export restrictions tightened. If V4 runs well on Huawei’s Ascend chips, it signals that domestic alternatives are getting real traction.

I’ve been watching the open-source vs. closed-source debate for years, and DeepSeek keeps proving that open models can punch above their weight. V3 was already impressive for its cost-efficiency—training it cost a fraction of what OpenAI or Google spend. V4 looks like it’s closing the performance gap even further.

Of course, we’re talking about a preview. The full release will tell us whether the benchmarks hold up in real-world use. But the timing is telling: exactly one year after DeepSeek rattled US rivals, they’re back with a model that’s explicitly designed to compete. That’s not a coincidence.

I’d love to see more independent benchmarks before getting too excited, but the direction is clear. DeepSeek isn’t just a footnote anymore—they’re a real contender. And if V4 delivers on its coding promises, the agent race just got a lot more interesting.