DeepSeek's 4h52m outage: chat and API down Thursday morning — and a full week's spend refunded as credit
DeepSeek went dark for nearly five hours on Thursday, October 15: its chat service and the APIs for DeepSeek V4 Pro and V4.1 Flash were down from 08:11 to 13:04 Beijing time. Once service was restored, DeepSeek posted a compensation notice promising every API account a full refund of its October 8–14 consumption — paid out as bonus credit.
What broke, and when
The outage was recorded on DeepSeek's official status page as "DeepSeek web/API partially unavailable," affecting three components: the chat service (对话服务), the DeepSeek V4 Pro API service, and the DeepSeek V4.1 Flash API service. The timeline, in UTC on October 15:
- 00:11 UTC (08:11 Beijing) — investigating: the issue is being triaged.
- 01:06 UTC — identified: the cause was found and a fix is being implemented.
- 04:49 UTC — monitoring: fix applied, results under observation.
- 05:04 UTC (13:04 Beijing) — resolved: the problem is fixed and service restored.
The page puts the total duration at 4 hours 52 minutes. Caixin reported that the web and app were showing "server temporarily unavailable" from around 8:00 Beijing time, and that the chat service was fully down from 9:08. For users on the US West Coast, the outage struck Wednesday evening — 17:11 to 22:04 on October 14 — squarely in prime coding-agent hours.
The make-good: a full week's spend back as credit
After service was restored, DeepSeek posted an apology and compensation notice on its API open platform. The terms, as reported by The Paper and IT之家: the refund equals what each account consumed from October 8 to October 14, returned in full as bonus credit (赠金), applied automatically. Future API billing is deducted from the bonus balance first — so the credit is spent before the user's own cash — and one detail worth checking on your balance page: the notice reportedly doesn't say whether the credit expires.
The credit tracks past spend, not actual loss: heavy users get a lot back, light users get little, even if the outage hit their launch hardest. Some users posted screenshots of refunds exceeding 1,000 yuan (about $150), per BigGo Finance. What the credit doesn't cover is the real cost — the requests your product failed to answer, the retries that piled up, and the support tickets that followed.
Not the first incident this year
Caixin counts 13 faults since September — most degraded performance on web and API, four involving partial outages of web and API. DeepSeek's own status page shows uptime for July to October 2026 of 99.74% for the V4 Pro API service, 99.46% for V4.1 Flash, and 99.45% for chat. DeepSeek has not said what technically failed; the compensation notice says the company has started a full review of service stability and will remove single points of dependency. When one incident takes down a chat product and two API models together, that wording reads as the diagnosis: something underneath was shared.
The lesson is provider-agnostic. A model API is one more dependency in the stack, and a week of free tokens doesn't keep your app answering while it's down. The practical move for anyone building on DeepSeek: point your SDK at a second OpenAI-compatible endpoint behind a failover that triggers on 5xx, 429s, and connection errors — and test it in staging before you need it in production.
Sources: DeepSeek status page — incident "DeepSeek web/API partially unavailable" (Oct 15); South China Morning Post: DeepSeek offers user refunds after suffering another major outage (Oct 15); Status Is Down: DeepSeek restores chat and API services after nearly five-hour outage (Oct 15); Gabriel Anhaia on DEV: DeepSeek's API was partly down for 4h52m (Oct 15).