DeepSeek, the Chinese artificial intelligence startup that helped roil world markets last year after it released its reasoning model R1, rolled out preview versions of its next major update on Friday, as competition between Chinese and U.S. AI firms intensifies. The company presented its V4 launch to users and developers who want to see how it stacks up against leading assistants and rival “reasoning” systems from companies such as OpenAI and Google, as well as Claude from Anthropic. DeepSeek said it is offering the V4 models in “pro” and “flash” versions.
DeepSeek said the V4 “pro” offering brings improvements in knowledge, reasoning and what it described as “agentic” capabilities—its ability to carry out complex tasks and workflows on its own. The company framed the release as an update to its earlier V3 model, which it released in late 2024, and positioned the V4 rollout as the latest step in a rapid cycle of model development.
The company also described a major technical change for both the “pro” and “flash” editions: a 1 million token context window. DeepSeek said that represents a significant step up from what it said was available in V3, which supported a 128,000 token context window, a change that is intended to let models process and recall more information in a single run.
In performance comparisons, DeepSeek said the V4 “pro Max” version delivered “superior performance” on standard reasoning benchmarks relative to OpenAI’s GPT-5.2 model and Google’s Gemini 3.0-Pro. It said V4 fell “marginally” short of OpenAI’s GPT-5.4 and of Gemini 3.1-Pro. DeepSeek also said its V4 “pro” version could outperform Claude’s Sonnet 4.5 and approach the level of Claude’s Opus 4.5 based on its own evaluation.
Alongside the stronger “pro” tier, DeepSeek said its “flash” version runs on simple agent tasks at a level it described as on a par with the “pro” edition, while keeping reasoning capabilities closely approaching it. DeepSeek also offered a free-to-use web and mobile chatbot, and it said developers can access and modify the V4 “open-source” models, describing that access as enabling changes and further building on its core technology.
Huawei also weighed in during the rollout, saying its Ascend chips and related technology are compatible with DeepSeek’s V4 models. Huawei’s statement was framed as evidence that it is possible to operate outside the Nvidia-dominated computing ecosystem amid what DeepSeek-adjacent commentary described as ongoing technological decoupling between China and the U.S.
The V4 launch landed on Thursday’s heels for another major AI release in the U.S.: DeepSeek’s rollout came hours after OpenAI released its GPT-5.5 model. The timing underscored how quickly each side is attempting to maintain momentum as users compare outputs and as investors track progress in model capabilities.
The rollout also arrived in a broader, ongoing dispute about how AI capabilities transfer between firms. In February, Anthropic accused DeepSeek and two other China-based AI laboratories of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities,” saying they did so using a technique called distillation that it described as training a less capable model on the outputs of a stronger one. OpenAI made similar allegations in a letter to U.S. lawmakers, and this week Michael Kratsios, chief science and technology adviser to U.S. President Donald Trump, accused foreign tech companies “principally based in China” of distilling leading U.S. AI systems and “exploiting American expertise and innovation.” China’s embassy in Washington responded by describing the allegations as “unjustified suppression of Chinese companies by the U.S.”
Even as it compared the V4 update to U.S. systems, some analysts said independent evaluations still matter. Lian Jye Su, chief analyst at technology research and advisory group Omdia, said, “Based on the benchmark results, it does appear DeepSeek V4 is going to be very competitive against its U.S. rivals.” Marina Zhang, an associate professor at the University of Technology Sydney, called the rollout a “pivotal milestone for China’s AI industry,” particularly as global competition intensifies around self-reliance in critical technologies. Ivan Su, a senior equity analyst at Morningstar, said the V4 update is a “competent” follow-up but not as large a breakthrough as the earlier R1 rollout, adding that “independent evaluations are needed before final conclusions can be drawn.”