Chinese artificial intelligence startup DeepSeek has unveiled a preview version of its highly anticipated V4 large language model, providing users with a chance to explore its innovative features. Marked by its debut on Friday, this release follows over a year after the company’s groundbreaking R1 reasoning model, which disrupted global tech markets with its impressive performance and cost-effectiveness.
Continuing its trend of transparency, DeepSeek’s latest model is open-source, enabling developers to download, run, and customize the code locally. The V4 model comes in two versions—“pro” and “flash”—designed to cater to different size requirements. DeepSeek asserts that V4 delivers impressive performance compared to domestic rivals, particularly excelling in agent-based tasks, knowledge processing, and inference.
Neil Shah, vice president of research at Counterpoint Research, remarked, “DeepSeek’s V4 preview is a serious flex,” highlighting its reduced inference costs compared to its predecessors. Inference costs encompass the computational and monetary resources needed to operate an AI model and generate results. Additionally, V4 has been fine-tuned for integration with leading agent tools such as Anthropic’s Claude Code and OpenClaw. Wei Sun, Counterpoint’s principal AI analyst, noted that V4’s benchmark profile indicates it could offer “excellent agent capability at significantly lower cost.”
DeepSeek, founded in 2023, first caught attention in late 2024 with its free, open-source V3 model, which claimed to be trained on less powerful chips and at a fraction of the cost of models from giants like OpenAI and Google. Its R1 reasoning model, launched in January 2025, set a new standard, outperforming many established large language models (LLMs) while raising eyebrows among investors about the sustainability of the U.S. lead in AI technology.
Despite its notable trajectory, analysts predict V4 will not have the same market impact as R1, as stakeholders have adjusted to the reality that Chinese AI solutions are both competitive and economical. Ivan Su, senior equity analyst at Morningstar, noted, “This is a framing that didn’t exist with R1, and that alone tells you how much domestic competition has intensified.” Indeed, since the introduction of R1, DeepSeek has encountered intensified competition from other players like Alibaba and ByteDance, all launching new models this year.
The context around the V4 release also raises questions regarding the chips used for its training. Huawei confirmed that its latest AI computing cluster, powered by Ascend AI processors, can support the V4 model. However, it remains uncertain how significantly Huawei’s hardware contributed to its training compared to the American AI chip leader Nvidia. U.S. export controls have restricted Chinese developers from easily acquiring Nvidia’s top-tier chips, prompting Beijing to bolster its domestic chip industry.
Counterpoint’s Wei Sun emphasized that V4’s capability to operate on local chips could prove pivotal in enhancing China’s AI self-sufficiency and decreasing reliance on Nvidia, contributing to a faster pace of global AI development. Following the V4 announcement, shares of Chinese contract chip manufacturers surged, with SMIC and Hua Hong Semiconductor jumping 9% and 15%, respectively.
As DeepSeek continues its rapid ascent in the burgeoning AI sector, the question remains: Will it surprise the world once more with its innovative capabilities? Only time will tell, but the landscape of AI technology is increasingly competitive and dynamic.
Original Source: https://www.cnbc.com/2026/04/24/deepseek-v4-llm-preview-open-source-ai-competition-china.html
Category :
Tags:
Publish Date: 2026-04-24 13:31:00
