China’s DeepSeek launched preview versions of its long-awaited V4 model, breaking months of silence from one of the country’s most closely watched AI labs.
While DeepSeek repeatedly delayed releasing a major model update, domestic rivals including Moonshot AI’s Kimi, MiniMax, Alibaba Group and ByteDance have aggressively pushed out updates.
V4-Pro’s agent ability has significantly improved from previous models, the company said on its official WeChat account on Friday. The model is now the “go-to agentic coding model” internally, with feedback showing that it beats Anthropic’s Sonnet 4.5 in user experience and delivering output quality closer to Opus 4.6’s non-thought mode, though it still lags behind Opus 4.6’s thought mode, they note.
The release ends a drawn-out wait for a major update model from DeepSeek. While the company repeatedly delayed releasing a major model update, domestic rivals including Moonshot AI’s Kimi, MiniMax, Alibaba Group and ByteDance have aggressively pushed out updates.
The Chinese company’s V4 model also marks a milestone for domestic chips.
DeepSeek said it validated one of the V4’s key efficiency techniques on both Nvidia GPUs and Huawei’s Ascend NPUs. Huawei said in a WeChat post that its entire Ascend line now offers full-stack support for DeepSeek V4 models.
The V4 model also uses “Sparse Attention,” a technique unveiled last year that enables the model to focus only on the most relevant parts rather than processing everything at once. That enables the model to handle much longer documents, the company said.
While V4-Pro is significantly more expensive than DeepSeek’s previous models, it remains much cheaper than its Western competitors. Anthropic, for example, charges $25 per million output tokens for its Opus 4.6 model, while 1 million output tokens for V4-Pro would cost $3.48.
DeepSeek also launched V4-Flash, a cheaper and faster version that holds its own against the V4-Pro on simpler tasks but trails on more demanding ones.
The startup is seeking at least $300 million in its first external fundraising, and investors have said that the company’s valuation would be pegged to the latest models’ performance, The Wall Street Journal reported.
