DeepSeek is back with V4: what the new model means for AI race

Chinese artificial intelligence startup DeepSeek began previewing its fourth-generation model on Friday.

The preview marks its most significant product update since the low-cost V3 system helped propel the company into the global spotlight last year.

The new model, DeepSeek V4, is being introduced in pro and flash versions and will succeed V3, which was released in December 2024, the company said in a post on its WeChat account.

The launch brings DeepSeek back to the centre of the fast-moving AI race at a time when investors and developers are watching Chinese model makers more closely for signs of technical progress and commercial ambition.

DeepSeek did not disclose pricing, performance benchmarks beyond broad claims, or a final release timetable for the full version.

Instead, the company described the preview as an early-stage rollout aimed at gathering real-world feedback before wider deployment.

That measured approach suggests DeepSeek is seeking to test demand and fine-tune the product before committing to a broader launch.

Upgrade follows V3 breakthrough

The new release represents the clearest sign yet that DeepSeek is trying to build on the momentum created by V3, the model that drew global attention for delivering strong performance at relatively low cost.

The startup shot to prominence last year after that breakthrough, which helped position it as one of China’s most closely watched AI challengers.

In its latest statement, DeepSeek described V4 as a major upgrade and the superior successor to V3.

The company’s choice to offer both pro and flash versions points to a familiar strategy in the AI sector: one model designed for higher-end use cases and another aimed at lighter, faster and potentially cheaper deployment.

While DeepSeek has not yet spelled out the exact differences between the two, the product split indicates an effort to reach a broader range of developers and enterprise customers.

Huawei link remains under watch

DeepSeek’s new model is also being closely followed because of its reported hardware ties.

Earlier this month, The Information report said that V4 would run on Huawei’s latest chips.

This is a notable development as Chinese AI firms adapt to US export controls and seek domestic alternatives to Nvidia hardware.

That makes the rollout significant not only as a product launch, but also as a test of how far China’s AI ecosystem can scale advanced models using locally sourced infrastructure.

For now, DeepSeek appears to be limiting what it reveals.

The company announced the preview on WeChat and has yet to offer a fuller technical breakdown or firm commercial roadmap.

Even so, the debut of V4 is likely to sharpen attention on the startup’s next steps, especially after a year in which it moved from relative obscurity to becoming one of the most discussed names in artificial intelligence.

Market interest builds again

The timing of the preview is notable.

DeepSeek is seeking to build on the attention won by its earlier models as the company re-enters the market with a fresh line-up and renewed visibility.

With competition intensifying between Chinese and US AI developers, V4 will be watched as a gauge of whether DeepSeek can turn last year’s viral rise into a more durable product and business franchise.