September 20, 2026
DeepSeek chat does not respond to users in Russia: over 6,000 complaints in one day
On September 20, DownDetector logged more than 6,000 reports about DeepSeek from Russia: everything is green on the service's status page.
“Server busy” instead of an answer, a silent chat, responses worse than usual. That was what DeepSeek looked like in Russia on September 20.
The outage covered 74 regions: the Sboy.rf tracker counted 2032 reports over 24 hours, versus 1781 the day before. Moscow accounted for 43% of complaints, Novosibirsk Oblast for 8%, and St. Petersburg for 7%.
No one acknowledged the outage. On DeepSeek's status page, every component is marked operational, uptime is 99.79–100%, and the September incident calendar is empty. Third-party monitor StatusGator counted 5 outages over 30 days and gave the page's accuracy a C rating. It recognizes problems with a delay of 30–120 minutes.
You can work around the chat. Previously, the official site was the only free gateway to the model. Now Yandex AI Studio offers DeepSeek V4 Flash through an OpenAI-compatible API with a 1 048 576-token window, while Cloud.ru has opened V4 to every registered user in Evolution Foundation Models: V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion, both with context windows of up to 1 million tokens.
The same model can be launched through Ollama with the single command `ollama run deepseek-v4-flash:cloud`: 304 billion parameters and a million-token context window, $0.22 per million input tokens off-peak and $0.44 at peak. For fully offline use, LM Studio helps: the 1.5-billion-parameter build weighs 1 GB, while 8–14 billion is used with 8 GB of memory.
Even on normal days, the free chat returns “server busy” most often from 4:00 to 7:00 and from 9:00 to 13:00 Moscow time.
In September, StatusGator has already counted three outage days for DeepSeek: September 2, 15, and 20.
Source
