DeepSeek V4 Preview Is Here, and It’s Aiming Right at OpenAI and Google

DeepSeek just released a preview of V4, its next-generation AI model, and the timing is no accident. It’s been almost exactly a year since their last model rattled the US AI industry, and this time they’re making a point of saying V4 can go toe-to-toe with closed-source systems from Anthropic, Google, and OpenAI.

That’s a bold claim, but the company isn’t just talking. They’re releasing V4 as an open-source model, which means anyone can poke around under the hood. That alone sets it apart from the increasingly walled gardens of GPT-4o, Gemini, and Claude.

The biggest improvement they’re touting is in coding. DeepSeek says V4 is significantly better than previous versions at writing and debugging code, which is the kind of capability that’s been driving the success of tools like ChatGPT Codex and Claude Code. If you’ve been using AI agents for development, you know how central that skill has become.

But here’s the part that caught my attention: DeepSeek explicitly highlights compatibility with domestic Huawei chips. That’s a big deal for China’s semiconductor industry, which has been scrambling to build alternatives to NVIDIA hardware under US export restrictions. It’s also a signal that DeepSeek is positioning itself as a patriotic tech champion, not just a scrappy startup.

I’ll be honest — I’m skeptical about some of the benchmarks. Chinese AI companies have a history of cherry-picking metrics, and independent verification is still thin. But the open-source nature of V4 means we’ll get real-world testing soon enough. If the coding claims hold up, this could be a genuine competitor for developer workflows.

The Verge has the full story, but the takeaway is clear: the gap between US and Chinese AI models isn’t closing as fast as some predicted, but DeepSeek is making sure it doesn’t widen either. V4 preview is available now, and I’m already downloading it to see how it handles my Python mess.