China’s DeepSeek suffers uncommon outage lasting a number of hours

The prolonged downtime is uncommon for DeepSeek, which has saved its companies up with out many glitches for over a yr.

China’s AI darling DeepSeek lately suffered one among its longest outages because the viral rise of its R1 and V3 fashions early final yr.

The beginning-up’s standing web page famous an preliminary incident at round 9.30pm native time (2.30pm GMT+1) on Sunday (29 March), which took practically two hours to resolve. Subsequently, within the early hours of right now (30 March), a unique incident was flagged, which took till 10.33am native time (3.30am GMT+1) to be absolutely resolved.

DeepSeek stated the incidents took greater than eight hours to be resolved, marking one of many longest outages to have an effect on the corporate’s internet chat companies. The reason for the outage stays unclear.

The prolonged downtime is uncommon for DeepSeek, which has saved its companies up with out many glitches for over a yr. Within the preliminary days after launching R1, DeepSeek’s API companies suffered a number of outages, quite a lot of which lasted an entire day. General, although, the platform has saved its companies operating 100pc of the time for a number of months in a row.

A yr after taking the AI business by storm with its giant language mannequin, DeepSeek is readying itself for a serious replace with V4. In accordance with reviews, the brand new V4 is a multimodal mannequin with image, video and text-generation options. V4 was reportedly educated with the assistance of Chinese language AI chipmakers Huawei and Cambricon.

Rival Chinese language AI corporations have launched quite a lot of new fashions in latest months in anticipation of DeepSeek, whose new mannequin may outperform ChatGPT and Claude, notably on duties that contain lengthy coding prompts. V4 is predicted to the corporate’s most necessary product launch since V3 and R1.

In February, Alibaba unveiled its newest AI mannequin, Qwen3.5, whereas ByteDance launched its fashionable Seedance 2.0, Zhipu unveiled GLM-5, MiniMax launched M2.5 and Moonshot launched Kimi K2.5.

Final September, DeepSeek launched V3.2-Exp, an experimental mannequin meant to be an “intermediate step” towards its next-generation structure. The V3.2-Exp mannequin’s efficiency is akin to the earlier V3.1 mannequin, performing marginally worse on some benchmarks.

Don’t miss out on the data that you must succeed. Join the Day by day Temporary, Silicon Republic’s digest of need-to-know sci-tech information.