DeepSeek Down for 5 Hours: API Users Get a Week Refunded

If your app called DeepSeek this morning and got nothing back, it was not your code. With DeepSeek down for close to five hours on 15 October 2026, the company chose an unusual fix for the damage: instead of crediting the hours lost, it is handing every API customer back a full week of spending.

Key points

• DeepSeek's status page logs the incident from 00:11 to 05:04 UTC, 4 hours 52 minutes, covering the DeepSeek V4 Pro and V4.1 Flash APIs and the web chat. In India that was 05:41 to 10:34 IST.
• API accounts get their spending from 8 to 14 October back as credit, added automatically and used before any new charges.
• DeepSeek has apologised and promised to remove single points of dependency. It has not said what failed.

DeepSeek down: the timeline and what broke

The status page tells the morning in four steps. Engineers began investigating at 00:11 UTC, identified the cause at 01:06, had a fix in place by 04:49 and declared the incident resolved at 05:04. The listed components are the V4 Pro and V4.1 Flash API services, and the incident title covers the web chat as well. According to OSCHINA's account, chat users saw a "server temporarily unavailable" message while it lasted.

It was the longest of several incidents this month. The same status page shows degraded performance on 1 October for an hour, on 8 October for three minutes, and two short web and API blips on 12 October of 11 and 15 minutes. On the morning we read it, the page put chat uptime since July at 99.44% and V4.1 Flash API uptime at 99.46%, below the V4 Pro API's 99.74%.

The refund: a week of DeepSeek API spending

DeepSeek posted the compensation notice on its developer platform, and outlets including OSCHINA reported its terms. Every API account receives credit equal to what it spent between 8 and 14 October. The credit arrives without a claim form, and future API charges come out of it first. In practice, a week of usage becomes free.

The amounts vary with how much each account used. Users quoted by Phoenix News and cited by OSCHINA reported refunds from just over 200 yuan to about 3,000 yuan. DeepSeek's apology was short: "We sincerely apologize for the inconvenience and trouble this outage has caused you," followed by a pledge to review the service and strip out single points of dependency so the same failure cannot recur.

Why the timing cost more than five hours

DeepSeek charges double during peak hours. Its pricing page defines peak as 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, with every other hour at half price. The outage ran from 00:11 to 05:04 UTC, so it swallowed the entire morning peak window, the three hours when DeepSeek's paying customers in China are busiest and when it earns the most per token.

Model	Input, cache miss (off-peak / peak)	Output (off-peak / peak)
DeepSeek V4.1 Flash	$0.15 / $0.30 per 1M tokens	$0.60 / $1.20 per 1M tokens
DeepSeek V4 Pro	$0.66 / $1.32 per 1M tokens	$1.98 / $3.96 per 1M tokens

That price list is why so many developers route traffic to DeepSeek, and why a morning without it hurts. Refunding a week rather than five hours is a commercial decision as much as an apology: it keeps price-sensitive customers from moving their traffic elsewhere.

The bigger DeepSeek picture this month

The outage lands in the middle of a fundraising push. CNBC reported on 6 October that DeepSeek was considering expanding its current round to as much as 100 billion yuan, about $14.9 billion, at a valuation near 500 billion yuan, with Bloomberg reporting an initial public offering targeted for early 2027. Investors pricing a company at that level will read the incident log as closely as the benchmark charts.

What to do if you build on DeepSeek

• Check your balance. The credit should already sit in your account on DeepSeek's developer platform, and it will absorb your next charges automatically.
• Add a fallback. A five-hour gap is long enough to break a product. Route failed calls to a second provider such as Claude, ChatGPT or Gemini, and test that path before you need it.
• Shift batch jobs off-peak. Work that can wait runs at half price outside DeepSeek's peak windows, and it is less exposed to the busiest hours.
• Watch the status page. Subscribe to status.deepseek.com so the next incident reaches you before your users do.

For a broader view of the assistants people switch to when one goes down, see our ranking of the best AI chatbots and the head-to-head ChatGPT vs Claude.

FAQ

Is DeepSeek down right now?

The 15 October outage was resolved at 05:04 UTC, 10:34 IST. For live status, check status.deepseek.com, which lists each component and every incident since July.

How much is the DeepSeek refund?

It equals each API account's own spending from 8 to 14 October, so it differs by account. Users have reported amounts from about 200 yuan to about 3,000 yuan.

Do I need to claim the DeepSeek credit?

No. Per the notice, the credit is added automatically, and future API charges are deducted from it before your paid balance.

Why was DeepSeek down?

DeepSeek has not disclosed the cause. Its status page says only that the cause was identified about an hour into the incident, and its notice promises to remove single points of dependency.

Does the refund cover DeepSeek chat users?

The notice covers API accounts and their spending. It does not mention compensation for people who were using the web chat during the outage.