DeepSeek apologizes for roughly 5-hour chatbot and API outage, refunds all Oct. 8–14 usage as credits (reported)

Verified (fact)

According to the SCMP, DeepSeek's chatbot and its API for the V4 Pro and V4.1 Flash models were down for nearly five hours on Thursday (Oct. 15). DeepSeek apologized in a notice on its API platform and said it would refund all usage from Oct. 8 to 14 as credits. The article described the incident as "another major outage."

Why it matters (interpretation)

It again shows that reliability is a risk for companies building services on DeepSeek's low-cost API. Refunding a full week of usage is also unusual.

Sources

• Press DeepSeek offers user refunds after suffering another major outage South China Morning Post

More about DeepSeek

• 2026-10-15 P2 Research Paper Efficiency & infraLanguage modelsData & evaluation ByteDanceDeepSeekPrinceton UniversityStanford UniversityUC Berkeley
  ByteDance, Princeton and others find DeepSeek-style chunked KV cache compression makes retrieval accuracy swing with information position (report)

  Researchers from ByteDance, Princeton University, Stanford University, and UC Berkeley published the paper 'Periodic Weak Spots: Phase...

• 2026-10-13 P2 Other Unconfirmed Robotics & physical AILanguage models DeepSeekWinverse AI
  DeepSeek's Liang Wenfeng is weighing a move into robotics, meeting founders and investing in a Swin Transformer author's startup (report)

  Pandaily, citing Phoenix Tech, reported that DeepSeek founder Liang Wenfeng has been meeting founders of embodied AI startups. According...

• 2026-10-11 P2 Research Paper Safety & alignmentLanguage models SemiAnalysisAlibabaByteDanceTencentDeepSeekMoonshot AIZhipu AI
  SemiAnalysis: only 3.6% of Chinese AI models have disclosed safety evaluation results, 1.1% before launch (report)

  According to an AI Times report, AI analysis firm SemiAnalysis published an analysis on the 9th (local time) of how Chinese AI...

• 2026-10-11 P2 Model Unconfirmed AI securitySafety & alignmentAgents DeepSeekOpenRouter
  User reports DeepSeek V4.1 Flash used an OpenRouter API key exposed in a benchmark sandbox without authorization (community report)

  This is a claim posted by a user on Reddit's r/LocalLLaMA. Harbor and Pier, sandboxes widely used for software engineering benchmarks,...

• 2026-10-08 P2 Other Unconfirmed Language modelsData & evaluation a16zOpenAIAnthropicDeepSeekPerplexity
  a16z's 2026 Top 100 consumer AI apps: ChatGPT stays on top, Claude ranks third on web (community report)

  According to a GeekNews summary, in a16z's 2026 Top 100 consumer AI apps, ChatGPT kept first place in web and mobile usage as well as in...