DeepSeek users hit limits on inactive chats and daily usage
Reports of blocked long-running conversations expose a gap between free AI access and dependable daily use.
Some DeepSeek users are returning to a long-running conversation only to find they cannot continue it. On November 16, China's STAR Market Daily reported that the app and website were restricting lengthy chats that had been inactive for more than seven days, with notices citing insufficient computing capacity. For someone treating a chat as a continuing work notebook, that interruption affects more than the next question: it breaks the routine of picking up an existing discussion.
Reports concern the consumer chat service. ITHome said feedback began arriving on November 15; its story appeared on the site's daily ranking the next day. Yicai described users who exhausted their allowance in less than two hours, and others who wanted to pay for reliable access but could not find that option in the app. Hitting an unexpected ceiling is different from planning work around an allowance stated in advance.
STAR Market Daily identified three constraints: inactive long conversations, a single-input limit of roughly 40,000 Chinese characters, and a daily usage ceiling. For scale, 40,000 characters is about twenty documents of 2,000 characters each. That describes one input, not a daily question allowance. Yicai said DeepSeek had not issued a consolidated explanation of the limits or a timetable for lifting them when its report was filed.
A long chat resembles meeting minutes that keep gaining pages: answering a new question involves handling relevant earlier material too. DeepSeek's developer documentation describes a cache that stores repeated opening content so it can be reused. Reuse reduces duplicate processing, while generating the next answer still requires computation. That helps explain why conversation history has a resource cost. Processing reused material and producing a new answer are different parts of that cost.
Developers face a separate billing structure. DeepSeek's official pricing charges for input and output text units and distinguishes cached from uncached input. Programs can access the model through that interface; ordinary app users want to reopen a conversation and carry on. The immediate tension is between a service woven into people's daily work and rules they struggle to predict. For heavy users, a dependable allowance now matters alongside the quality of the answer.
Original source published: STAR Market Daily, November 16, 2026; ITHome, November 16, 2026, 11:44 Beijing time / November 15, 2026, 10:44 PM EST.
Share this article
What do you think of this story?
Anonymous feedback, up to 5 reactions per article.
Reader comments
Comment without signing in, or sign in to display your member name.
Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.
No comments yet. Be the first to share.