China’s DeepSeek unveils much-awaited V4 AI mannequin amid rising US-China tech rivalry

Hong Kong

Chinese synthetic intelligence startup DeepSeek on Friday launched preview variations of its newest main AI mannequin, V4, as competitors between China and the United States in superior AI applied sciences intensifies.

The launch had been eagerly awaited by customers and trade watchers trying to evaluate the mannequin with main American rivals equivalent to OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini.

DeepSeek mentioned its open-source V4 household consists of “Pro” and “Flash” variants, that includes main upgrades in data, reasoning, and “agentic” talents — that means the capability to independently full advanced duties and workflows.

The V4 mannequin succeeds DeepSeek’s V3 system, launched in late 2024. However, it was the corporate’s reasoning-focused mannequin DeepSeek R1 that drew world consideration in January 2025 after claims it delivered sturdy efficiency at decrease value than comparable Western fashions.

According to the corporate, its “V4 Pro Max” model performs higher than GPT-5.2 and Gemini 3.0-Pro on a number of reasoning benchmarks, whereas trailing barely behind newer variations GPT-5.4 and Gemini 3.1-Pro.

DeepSeek additionally mentioned the V4 Pro mannequin outperforms Claude Sonnet 4.5 in autonomous job execution and approaches the extent of Claude Opus 4.5 based mostly on inside testing.

Technology analyst Lian Jye Su of Omdia mentioned benchmark knowledge suggests the brand new mannequin may very well be extremely aggressive in opposition to US rivals.

Marina Zhang of University of Technology Sydney described the rollout as an necessary second for China’s AI sector, particularly as nations search larger self-reliance in strategic applied sciences.

DeepSeek provides free internet and cellular chatbot companies and promotes its programs as open supply, permitting builders to switch and construct on its underlying expertise. The firm mentioned each V4 variants help a 1 million-token context window, a pointy soar from V3’s 128,000-token restrict.

Still, some consultants urged warning. Ivan Su of Morningstar mentioned V4 seems strong however might not be as disruptive as the sooner R1 launch, noting that stronger impartial testing is required.

READ MORE: Assam-born Fazlur Rahman finds most cancers’s hyperlink with atmosphere

The replace additionally comes amid persevering with tensions over allegations from OpenAI and Anthropic that Chinese companies used “distillation” strategies to be taught from US fashions. China has rejected these claims and accused the US of unfairly focusing on Chinese firms.