Back to Flash News

DeepSeek imposes limits on free users due to insufficient computing power; both long conversations and daily quotas are restricted, prompting a wave of complaints from emotional companionship users

Around November 16, 2026, DeepSeek’s app and web version for regular users began imposing usage limits, with the system citing “insufficient computing power” as the reason. Specific measures include no longer supporting long conversations with no active interaction for more than 7 days (some users reported 72 hours), with a suggestion to start a new conversation or export chat history; the maximum input per session has been reduced to about 40,000 Chinese characters; and a new daily conversation quota has been added, after which users are prompted that, “due to server congestion, the usage quota has reached its limit.” Some users reported hitting the limit after less than two hours of continuous use. As of now, DeepSeek has not issued a unified official notice on the quota rules, nor has it explained when the restrictions will be lifted.

The restrictions mainly target consumer-facing free services, and the API side is not affected. The background is surging demand: according to OpenRouter data, DeepSeek V4.1 Flash processed about 59.2 trillion to 61.2 trillion tokens in the week from November 9 to 15, ranking first globally, and its weekly usage nearly tripled within two months; the official status page shows that there have been multiple service outages since October.

User reactions have been concentrated: official comment sections and app store reviews have been flooded with many dissatisfied messages. Many heavy users use DeepSeek for role-playing, interactive fiction, and emotional companionship scenarios, and are highly dependent on long conversations; after the interruption, they showed obvious emotional reactions, with some comments bluntly saying they had “broken up” or that their “husband is gone,” and expressing hope to obtain higher quotas through payment.

Source: reports from 科创板日报, 第一财经, 澎湃新闻, IT之家, 中国青年报 and others on November 16-17; public OpenRouter data.

ABAB AI Insight

The core issue is active resource reallocation in the face of explosive demand that is outpacing supply: long conversations continuously occupy KV Cache (GPU memory), and clearing idle sessions can prioritize active inference. The official side has not formally confirmed that “insufficient computing power” is the only reason, nor has it given a recovery timeline; this is a silent product-side adjustment. The reactions from emotional companionship users are real and concentrated, but DeepSeek itself is not primarily positioned as a product for that scenario.