X

DeepSeek V3 Is Now Reportedly the Best Nonreasoning AI Model

According to a new intelligence index analysis, DeepSeek V3 is outcompeting GPT-4.5, Grok 3 and Gemini 2.0.

Headshot of Imad Khan
Headshot of Imad Khan
Imad Khan Senior Reporter
Imad is a senior reporter covering Google and internet culture. Hailing from Texas, Imad started his journalism career in 2013 and has amassed bylines with The New York Times, The Washington Post, ESPN, Tom's Guide and Wired, among others.
Expertise Google | AI | Internet Culture
Imad Khan
2 min read
DeepSeek AI logo on a phone
James Martin/CNET

DeepSeek V3 is now the best nonreasoning AI model on the market, besting OpenAI, Grok and Google, AI benchmarker Artificial Analysis posted Tuesday. 

While AI companies trade blows for the top spot as new updates are released, what makes DeepSeek V3's accomplishment noteworthy is that it's the first time an open-weights model is leading the pack. Artificial Analysis says its intelligence index measures responses in reasoning, knowledge, math and coding.

AI Atlas badge tag

Open weights means that DeepSeek V3 is fully open source and that the entire model is publicly available for users to tweak and modify. OpenAI, despite its name, isn't willing to give away all of ChatGPT's secrets. The AI company keeps a tight lid on how it trains its models and the underlying architecture of how it operates. There are merits to both the open and closed approach to software development, and the choice between the two often comes down to business interests. 

Given that V3 isn't a "reasoning" AI model, it isn't as powerful. At the same time, it's much faster, meaning it's cheaper and more feasible to run for most applications.

DeepSeek V3 launched last December and its latest update puts increased competition on American AI companies. DeepSeek gained prominence earlier this year when it released its R1 reasoning model for free. It was the first time a high-level reasoning model, one that recursively went back and checked its answers before giving a final output, was widely available to the masses. DeepSeek R1 also impressed with its lower cost to run, showing that the firm had made innovations in efficiency. 

But the world of reasoning AI models are better suited for research or tasks that require massive datasets. Nonreasoning models, like GPT-4.5 and Google Gemini 2.0, are better suited for most applications, as they're faster and cheaper to operate. DeepSeek V3's ascendance shows that Chinese AI firms can outcompete American companies while also keeping things open-source.

This threat is why OpenAI is pushing the Trump administration to lift restrictions on training with copyrighted material, arguing that if fair use limitations remain, American AI companies won't be able to compete against China.