HONG KONG -- DeepSeek, the Chinese artificial quality startup that shook satellite markets past year, launched preview versions of its latest large update Friday arsenic the AI rivalry betwixt China and the U.S. heats up.
DeepSeek’s V4 has been keenly anticipated by users keen to trial however it compares to U.S. competitors similar OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini. Anthropic and OpenAI person accused DeepSeek of unfairly gathering its exertion disconnected their own.
Some manufacture analysts had expected the caller exemplary to get much than a period earlier astatine the commencement of the Lunar New Year.
DeepSeek says the caller V4 open-source models, which see “pro” and “flash” versions, person large improvements successful knowledge, reasoning and successful their “agentic” capabilities – the quality to execute analyzable tasks and workflows autonomously.
V4 is simply a successor to V3, an AI exemplary that DeepSeek released successful precocious 2024.
But it was DeepSeek’s specialized “reasoning” AI model, called R1, that took markets by astonishment with its merchandise successful January 2025. DeepSeek claimed it was much cost-effective than OpenAI’s akin exemplary and it became a awesome of however China was catching up with the U.S. successful technological advancements.
DeepSeek said the “V4 Pro Max” mentation has “superior performance” successful presumption of modular reasoning benchmarks comparative to OpenAI’s GPT-5.2 exemplary and Google’s Gemini 3.0-Pro. It falls “marginally” abbreviated of GPT-5.4 and Gemini 3.1-Pro, it said.
In presumption of “agentic” capabilities, the Chinese institution said the V4 “pro” mentation could outperform Claude’s Sonnet 4.5 and approaches the level of Claude's Opus 4.5 exemplary based connected its ain evaluation.
The “flash” mentation of V4 performs connected a par with the “pro” mentation connected elemental cause tasks and has reasoning capabilities intimately approaching it, DeepSeek said.
“Based connected the benchmark results, it does look DeepSeek V4 is going to beryllium precise competitory against its U.S. rivals,” said Lian Jye Su, main expert astatine the exertion probe and advisory radical Omdia.
Marina Zhang, an subordinate prof astatine the University of Technology Sydney, said DeepSeek's V4 rollout is arsenic a “pivotal milestone for China’s AI industry”, particularly arsenic planetary contention intensifies successful the pursuit of self-reliance successful captious technologies.
DeepSeek offers a free‑to‑use web and mobile chatbot. Unlike the apical models from Anthropic, Google and OpenAI, it describes its exertion arsenic “open source” successful the mode that it enables developers entree to modify and physique connected its halfway technology.
Both the V4's “pro” and “flash” versions person a 1 cardinal token discourse window, a parameter of however overmuch accusation an AI exemplary tin process and recall, and tally connected a much businesslike basis, the startup said. That is simply a important betterment from before, since the V3 supported a 128,000 token discourse window.
A study from Microsoft successful January showed usage of DeepSeek has been gaining crushed successful galore processing nations.
However, immoderate analysts stay skeptical. Ivan Su, a elder equity expert astatine Morningstar, said portion V4 is simply a “competent” follow-up, it’s not arsenic large a breakthrough arsenic the rollout of R1.
“Domestic contention has intensified importantly since R1’s release,” Su said. “Against U.S. models, DeepSeek’s ain valuation suggests its capabilities mostly lucifer connected astir fronts, but autarkic evaluations are needed earlier last conclusions tin beryllium drawn.”
In February, Anthropic accused DeepSeek and 2 different China-based AI laboratories of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities to amended their ain models.” It said they did that utilizing a method called distillation that “involves grooming a little susceptible exemplary connected the outputs of a stronger one.” OpenAI made akin allegations successful a missive to U.S. lawmakers.
This week, Michael Kratsios, main subject and exertion advisor to U.S. President Donald Trump, besides accused overseas tech companies “principally based successful China” of distilling starring U.S. AI systems and “exploiting American expertise and innovation.”
China’s embassy successful Washington deed backmost astatine the allegations, describing them arsenic “unjustified suppression of Chinese companies by the U.S.”
___
O’Brien reported from Providence, Rhode Island.
English (CA) · 
English (US) · 
Spanish (MX) ·