DeepSeek Unveils Major Update Amidst Intensifying AI Rivalry
Hong Kong: DeepSeek, the Chinese artificial intelligence startup that disrupted global financial markets last year, has unveiled preview versions of its latest update on Friday.
This launch comes as the competitive landscape between China and the U.S. escalates in the realm of artificial intelligence.
The eagerly awaited V4 has drawn significant interest from users eager to evaluate its performance against American counterparts such as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini.
Notably, allegations have emerged from Anthropic and OpenAI, accusing DeepSeek of constructing its technology utilizing their own innovations unfairly.
Some analysts had anticipated the new model’s debut would occur over two months earlier, coinciding with the onset of the Lunar New Year.
DeepSeek asserts that the new open-source V4 models, including “pro” and “flash” versions, exhibit substantial enhancements in areas such as knowledge, reasoning, and “agentic” capabilities, allowing for the execution of complex tasks autonomously.
A significant shift in this iteration is its reliance on chips manufactured by Chinese tech behemoth Huawei, rather than American suppliers like Nvidia.
V4 succeeds the V3 model released by DeepSeek in late 2024.
However, it was the launch of DeepSeek’s specialized reasoning model, dubbed R1, that astonished markets when it debuted in January 2025.
The company claimed R1 was more cost-efficient than analogous models from OpenAI, symbolizing China’s rapid advancements in technology vis-à-vis the U.S.
In a proclamation, DeepSeek noted that the “V4 Pro Max” version exhibits “superior performance” in standard reasoning tests compared to OpenAI’s GPT-5.2 and Google’s Gemini 3.0-Pro models, falling just shy of GPT-5.4 and Gemini 3.1-Pro.
Notably, DeepSeek’s announcement followed closely on the heels of OpenAI’s release of its new GPT-5.5 model.
Addressing “agentic” functionalities, DeepSeek asserted that the V4 “pro” version may surpass Claude’s Sonnet 4.5 and closely rival Claude’s Opus 4.5, according to its evaluation metrics.
The “flash” variant of V4 performs comparably to the “pro” version in simpler agent tasks and has reasoning capabilities that closely align with it, as claimed by the company.
“Based on the benchmark results, it appears that DeepSeek V4 is poised to be highly competitive against its U.S. rivals,” stated Lian Jye Su, chief analyst at the technology research and advisory firm Omdia.
Marina Zhang, an associate professor at the University of Technology Sydney, remarked that the rollout of DeepSeek’s V4 represents a “crucial milestone for China’s AI industry,” particularly as the global race for technological self-reliance intensifies.
DeepSeek offers a freely accessible web and mobile chatbot. In contrast to leading models from Anthropic, Google, and OpenAI, it characterizes its technology as “open source,” granting developers the ability to modify and enhance its core frameworks.
Both V4’s “pro” and “flash” models feature a 1 million token context window, enhancing their capability to process and recall information, which signifies a remarkable advancement from V3, which had a mere 128,000 token context window.
Huawei released a statement on Friday affirming that its Ascend chips and associated technologies are compatible with DeepSeek’s V4 models.
This compatibility signifies the technical feasibility of operating outside the Nvidia-centric computing paradigm amidst ongoing technological decoupling between China and the U.S., as noted by Zhang.
A report from Microsoft in January indicated that DeepSeek’s usage is gaining traction in various developing nations, particularly in regions where Huawei smartphones are prevalent.
Nonetheless, some analysts express skepticism. Ivan Su, a senior equity analyst at Morningstar, commented that while V4 is a “competent” successor, it does not constitute a groundbreaking advance akin to R1’s debut.
“Domestic competition has significantly intensified since R1’s launch,” Su noted. “Against U.S. models, DeepSeek’s self-assessment suggests its capabilities largely match on many fronts; however, independent evaluations are essential for conclusive assessments.”
In February, Anthropic charged DeepSeek and two other Chinese AI laboratories with executing “industrial-scale campaigns” to “illicitly extract Claude’s capabilities to bolster their own models.”
They alleged that this was accomplished through a technique known as distillation, where a lesser-capable model is trained using the outputs of a more proficient one. OpenAI echoed similar accusations in a letter to U.S. lawmakers.
This week, Michael Kratsios, chief science and technology adviser to U.S. President Donald Trump, further accused foreign tech companies, “principally based in China,” of distilling leading U.S. AI systems and “exploiting American ingenuity and expertise.”
In response, China’s embassy in Washington labeled these allegations as “unjustified suppression of Chinese enterprises by the U.S.”
Source link: Republicworld.com.
