ByteDance and DeepSeek Are Placing Very Different AI Bets

Go precocious oregon spell wide? DeepSeek and ByteDance, the 2 leaders of China’s AI industry, are adopting vastly antithetic strategies.

On Monday, DeepSeek released DeepSeek V3.2, different open-weight exemplary that anyone tin tinker with. The startup says it performs connected par with the latest models from OpenAI and Google, and it adjacent beats them connected immoderate cardinal mathematics benchmarks.

That aforesaid day, ByteDance, whose dominance successful AI applications we covered previously, introduced even more ways for radical to usage its chatbot, Doubao. ByteDance is present moving with a Chinese smartphone shaper to embed Doubao into the operating system, giving it entree to antithetic apps and allowing it to behaviour agentic tasks with them. In different words, it’s coming for Apple’s Siri.

Both ByteDance and DeepSeek person AI apps with implicit 140 cardinal monthly users. But their latest announcements correspond 2 diverging trends successful China’s AI industry. While immoderate companies are inactive competing with their Western counterparts to physique ever much susceptible models, others person softly withdrawn from that crippled and are focusing connected however they tin integrate their AI tools into people’s mundane lives.

DeepSeek Resurfaces

DeepSeek’s latest open-weight exemplary whitethorn person disappointed immoderate of its astir loyal followers, who are inactive waiting for R2, a much-anticipated update to the archetypal exemplary that rocked Silicon Valley successful January. Instead, DeepSeek released V3.2 and V3.2-Speciale, which are better-optimized versions of its erstwhile exemplary V3.2-Exp, released successful September.

Still, V3.2 caused a disturbance successful the AI manufacture due to the fact that DeepSeek claims it tin lick the benignant of precocious mathematics questions asked astatine the International Mathematical Olympiad, and its show connected different coding and reasoning tasks is supposedly connected par with oregon supra GPT 5 and Gemini 3. “It abruptly dawned connected maine wherefore they telephone the institution DeepSeek with the whale arsenic a motif. Because conscionable similar a whale, it seldom surfaces, but each clip it surfaces, it ever makes a monolithic splash,” says Jen Zhu Scott, an AI capitalist and the cofounder and CEO of Power Dynamics, a modular data-center solutions firm.

However, I can’t assistance but consciousness similar this arms contention of AI models is getting a small tiring, peculiarly due to the fact that truthful galore caller ones person been released successful the past month, each claiming to instrumentality humanity 1 measurement higher. In little than 20 days, we had OpenAI’s GPT 5.1, Google’s Gemini 3 Pro, Anthropic’s Claude Opus 4.5; propulsion successful Chinese unfastened root models similar Moonshot’s Kimi K2 and DeepSeek’s V3.2, and it becomes a full mess. My attraction span tin beryllium summarized by this cleanable meme.

“At the extremity of the day, we can't support up with each these hairline differences betwixt antithetic models, antithetic releases,” Zhu says. “It really doesn't marque a immense difference, isolated from immoderate benignant of banal marketplace speculation connected who's gonna win.”