DeepSeek said it will return a full week of API spending to affected API users as promotional credits after a service outage on Oct. 15 disrupted its web interface, app chat, and API access for nearly five hours.
The company’s status page shows the incident started at 8:11 a.m. UTC+8 and was fully resolved at 1:04 p.m., for a total duration of 4 hours and 52 minutes. The report described it as DeepSeek’s longest outage since July.
What users will get back
DeepSeek said compensation will be calculated based on each account’s API spending from Oct. 8 to Oct. 14, covering the seven days before the outage. Charges incurred on Oct. 15, the day of the disruption, are not included.
The spending refers to fees deducted when users call the API, based on token usage multiplied by the model’s unit price. The refund will not be paid in cash. Instead, it will be issued as promotional credit, a separate balance inside a DeepSeek account that sits alongside prepaid funds. The credit is being deposited automatically, with no application required.
Chinese tech outlet Fast Technology reported on Oct. 15 that, after logging into the API developer site, it found the compensation had already arrived.
How the credit will be used
DeepSeek’s notice said that API billing will deduct from the promotional credit account first. Its pricing page also states that when both prepaid balance and gifted balance are available, the system prioritizes the gifted balance.
In practical terms, users’ future API calls will consume the newly issued credit before touching the money they previously topped up themselves.
Who is not covered
Free users on the web and app are not part of the compensation plan. Those users also lost access to chat services during the outage, but because they had no API spending to refund, they are receiving no compensation. The company’s notice did not set out any separate arrangement for them.
No public explanation for the outage
DeepSeek did not disclose the cause of the incident. Its notice said only that it had launched a full investigation and remediation process and would "comprehensively sort out and remove single-point dependencies."
The report noted that the status page showed web chat, the V4.1 Flash API, and the V4 Pro API all ran into problems at the same second, which is consistent with the idea of a shared component failing. DeepSeek did not identify that component, and the report said that point remains an inference rather than a confirmed explanation.
Status page history
From July 22 to Oct. 15, the status page listed 36 incidents in total, most of them classified as degraded performance and resolved within a few minutes to half an hour.
This 292-minute disruption was the longest over that period. The second-longest was a July 22 outage that made web search unavailable for 233 minutes and affected only search. The third-longest was a partial disruption involving the web service and API in the early hours of Sept. 15, lasting 180 minutes.
Since Sept. 2, the status page has recorded 13 incidents, including four partial outages. The first 12 incidents together added up to about 7 hours and 57 minutes, while this single event alone lasted close to five hours.
According to BlockTempo, DeepSeek has run into some technical trouble after recently rolling out new models.