DeepSeek Hit by Its 18th Major Outage of the Year as Chat Service and V4.1 Flash API Go Down

DeepSeek suffered a broad service disruption on Thursday morning Beijing time, with its website and mobile apps displaying a "server temporarily unavailable" message from shortly after 8:00 AM. According to the company's official status page, the partial outage began at 8:11 AM Beijing time, affecting the Chat Service and the DeepSeek V4.1 Flash API, with degraded performance across both the web product and the API.

The status page initially said the team was "investigating the issue to identify the cause." A later update shifted the incident state to cause identified and fix in progress, though Chinese media testing found the web client still unreachable at 9:47 AM — more than 90 minutes after the first reports. A follow-up status update added the V4 Pro API to the list of affected components, according to a Sohu report, meaning three of DeepSeek's main service surfaces were degraded at once.

Third-party monitoring picked up the incident quickly. StatusGator recorded 86 user reports within 24 hours, with complaints arriving from Luxembourg, Poland and New York, suggesting the disruption was not regional. Independent outage tracker accounts on X also flagged the incident within minutes of the first user reports.

It is far from an isolated event. DeepSeek's chat service also suffered a partial interruption on October 12, and a tally circulating via Sina Tech counts this as at least the 18th major service disruption this year. The repeated incidents have a common thread: Chinese industry commentary increasingly attributes them to a structural compute imbalance, where user and API demand keeps compounding faster than the company's inference capacity can expand, and peak-hour concurrency pressure transmits directly into stability.

The outage inevitably set off the community's favorite ritual. Within the hour, users on Reddit and Chinese platforms were speculating that the downtime masked pre-launch maintenance for a new model, and a previously unknown account named DeepSeekHarness appeared on Bilibili at roughly the same time, feeding the rumor mill. There is no official connection between the outage and any model release, and the speculation should be treated as exactly that — speculation. DeepSeek has not announced a cause or a restoration timeline beyond the status page updates.

The timing is awkward for a company whose economics depend on extreme throughput. DeepSeek has built its reputation on API prices far below frontier-lab rates, and its V4.1 Flash tier in particular is pitched at high-volume, latency-sensitive agent workloads. Every reliability incident erodes the one advantage that cheap inference cannot buy back: the confidence of developers routing production traffic. Rivals have noticed — several Chinese cloud providers now advertise DeepSeek-compatible endpoints with enterprise SLAs as a differentiator.

For now, the practical lesson repeats what the October 12 incident already taught: teams running anything important on DeepSeek need a fallback route. If the pattern of roughly one major outage every few weeks holds, capacity — not capability — may become the binding constraint on DeepSeek's growth for the rest of the year.