HONG KONG (AP) — DeepSeek, the Chinese artificial intelligence startup that made waves in global markets last year, unveiled preview versions of its latest major update on Friday, amid intensifying AI competition between China and the U.S. The highly anticipated V4 models are expected to be tested by users eager to see how they stack up against U.S. competitors such as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini. However, both Anthropic and OpenAI have alleged that DeepSeek unfairly based its technology on their models.
Initially, industry analysts had forecasted the new model’s arrival over two months ago, coinciding with the Lunar New Year. DeepSeek asserts that the new V4 open-source models, available in “pro” and “flash” versions, feature significant advancements in knowledge, reasoning, and “agentic” capabilities, which refer to the AI’s ability to autonomously perform complex tasks. A notable shift in DeepSeek’s architecture includes reliance on computer chips from the Chinese tech giant Huawei, reducing dependency on U.S. manufacturers like Nvidia.
The V4 series follows the release of V3 in late 2024. However, it was the specialized reasoning AI model, R1, that captured market attention with its January 2025 launch, touted as more cost-effective than OpenAI’s equivalent. This model symbolized China’s progress in narrowing the technological gap with the U.S.
DeepSeek claims that the “V4 Pro Max” version demonstrates “superior performance” in standard reasoning benchmarks compared to OpenAI’s GPT-5.2 and Google’s Gemini 3.0-Pro, although it falls “marginally” short of GPT-5.4 and Gemini 3.1-Pro. The timing of DeepSeek’s release was particularly notable, coming just hours after OpenAI launched its own GPT-5.5 model.
In terms of agentic capabilities, the V4 “pro” version reportedly outperforms Claude’s Sonnet 4.5 and approaches the capabilities of Claude’s Opus 4.5 based on internal evaluations. Meanwhile, the “flash” version matches the pro version on simpler agent tasks and closely aligns in reasoning abilities. “Based on the benchmark results, it does appear DeepSeek V4 is going to be very competitive against its U.S. rivals,” remarked Lian Jye Su, chief analyst at the technology research firm Omdia.
Marina Zhang, an associate professor at the University of Technology Sydney, characterized the V4 rollout as a “pivotal milestone for China’s AI industry,” especially as the global competition for self-reliance in critical technologies intensifies. Unlike the leading models from Anthropic, Google, and OpenAI, DeepSeek’s offerings are labeled as “open source,” allowing developers to modify and build upon its core technology without barriers.
Both the V4 “pro” and “flash” versions have a context window of 1 million tokens, significantly improving from the 128,000-token limit seen in V3. This enhancement allows for better processing and recall of information. Huawei announced that its Ascend chips are compatible with DeepSeek’s V4 models, showcasing the technical feasibility of operating outside Nvidia’s dominant ecosystem amid ongoing technological decoupling between China and the U.S. A Microsoft report from January indicated that DeepSeek is gaining traction in several developing nations, particularly where Huawei phones are prevalent.
However, skepticism remains among some analysts. Ivan Su, a senior equity analyst at Morningstar, described V4 as a “competent” successor but noted that it does not represent as significant a breakthrough as R1. “Domestic competition has intensified significantly since R1’s release,” he stated. “While DeepSeek’s own evaluations suggest its capabilities largely match those of U.S. models, independent assessments are necessary for definitive conclusions.”
In February, Anthropic accused DeepSeek and two other Chinese AI labs of conducting “industrial-scale campaigns” to illicitly extract capabilities from Claude to improve their own models. This practice reportedly involves a technique known as distillation, where a less capable model is trained on the outputs of a more advanced one. OpenAI made similar allegations to U.S. lawmakers, while Michael Kratsios, chief science and technology adviser to U.S. President Donald Trump, accused foreign tech firms based in China of exploiting American innovations.
China’s embassy in Washington responded to these claims, labeling them as “unjustified suppression of Chinese companies by the U.S.” As the AI landscape continues to evolve rapidly, the implications of DeepSeek’s V4 launch could resonate far beyond national boundaries, reflecting a shifting balance of technological power.
See also
OpenAI Launches Images 2.0 with Enhanced Precision and Multi-Language Support
AI Detection Tools Emerge as Guardians of Integrity in Synthetic Media Creation
Midjourney Launches V8 Alpha, Enhancing Image and Video Generation with New Features
AI Therapist Risks Highlighted: OpenAI’s Data Policies Raise Privacy Concerns
ComfyUI Raises $30M at $500M Valuation, Enhancing AI Media Control for Creators
