Chinese AI startup DeepSeek has finally launched a preview version of its highly anticipated V4 model, putting an end to months of anticipation. This release comes a year after the controversial R1 release that caused a stir in the U.S. market and raised concerns among Silicon Valley AI companies.
The rollout of DeepSeek-V4 Preview signifies the company’s commitment to staying ahead in the frontier-model race. The open-source model is available in two series: V4 Flash and V4 Pro. DeepSeek claims that its V4 model “leads all current open models, trailing only Gemini-3.1-Pro.”
According to DeepSeek, the V4 model surpasses all current open models in Math, STEM, and Coding, rivaling even the top closed-source models. The V4 Pro series boasts enhanced agentic capabilities and rich world knowledge, positioning it as a strong contender in the AI landscape.
Industry experts have reacted positively to DeepSeek’s V4 preview, with Counterpoint Research Vice President Neil Shah calling it a “serious flex.” The model’s benchmark profile suggests that it could offer excellent agent capabilities at a lower cost, making it an attractive option for businesses.
While the debut of V4 may not have the same market impact as R1, it has already made waves in China. GPU and domestic chip stocks rallied following the release, indicating a positive reception in the market. The chip stack behind V4, including the hardware it runs on during inference, is a key question that remains unanswered.
Overall, DeepSeek’s V4 model represents a significant advancement in the field of AI, powered by Beijing’s growing domestic AI hardware ecosystem. With its impressive performance and cost-effectiveness, V4 is set to make a mark in the AI industry.
[Original article source: CNBC]