TOPICS 

    Subscribe to our newsletter

     By signing up, you agree to our Terms Of Use.

    FOLLOW US

    • About Us
    • |
    • Contribute
    • |
    • Contact Us
    • |
    • Sitemap
    封面
    NEWS

    Bot, Interrupted: DeepSeek Users Report New Limits on Free Chats

    Users are widely reporting that the model now caps input in a given chat — the first time the company has restricted use of its free AI model.

    Chinese AI model DeepSeek has reportedly begun limiting use of its free chatbot by reducing the number of characters users can input and discontinuing inactive conversations after seven days. 

    DeepSeek users began reporting seeing an error message — “Due to server overload, the use limit has been reached” — on Monday when they tried to continue chats inactive for a week or when they tried to input larger volumes of text, according to posts on microblogging platform Weibo.

    According to domestic media, users of the large language model (LLM) can now only enter around 40,000 characters at a time (a page of Chinese text typically ranges from 500 to 800 characters). Previously, the chatbot had no publicly stated text limit. Once users reach their daily limit, DeepSeek displays an error message.

    It is reportedly the first time DeepSeek, headquartered in China’s eastern city of Hangzhou, has restricted free-user access since its launch in January 2025. 

    According to its official status page, DeepSeek has experienced 57 outages or performance degradations this year, without providing additional explanation. Domestic media reported that disruptions in the first half of the year mainly involved slower responses, delays in data generation, and occasional request failures. After October, prolonged global outages began to occur more frequently.

    Usage caps apply only to individual users, but because DeepSeek has no related paid plan, those who hit their capacity cannot pay for more. App developers, who pay to use DeepSeek through an application programming interface, or API, a connection that shares data between their software and the AI model, are subject to separate limits and per-use billing.

    As of Wednesday, DeepSeek had not announced a reason for the new rules or whether it might lift them. Developers who spotted references to a “Coding Plan” in the website’s code have speculated that the company may soon implement a paid subscription plan for individual users.

    Online, users debated the new limits, with many expressing outrage and complaining that the company had not rolled out a paid plan option for individual users before the usage limit. “I feel like I’m a widow who just lost her husband,” wrote one user on lifestyle platform Xiaohongshu, also known as RedNote.

    Others sympathized with DeepSeek, arguing that data is costly or hypothesizing that the company was clamping down on users developing companion relationships with its AI, alluding to China’s ban on AI products designed to mimic human personalities that took effect on July 15.

    Data from OpenRouter, a U.S.-based platform that provides access to numerous AI models through one interface, showed DeepSeek’s latest model processing 59.2 trillion tokens the week of Nov. 9 to 15, topping OpenRouter’s weekly token usage chart.

    Long chats are more costly for LLM developers because models must parse more information with each reply. Industry watchers told domestic media that the higher costs mean DeepSeek may now have to choose between training new models, serving paying API customers, and providing free chats.

    DeepSeek founder Liang Wenfeng has long said the company has moved toward developing its artificial general intelligence capabilities — the drive to create an AI that can match or surpass humans across a wide range of tasks — rather than consumer revenue. 

    Earlier this year, Liang said computing power is the biggest factor holding the company back. He also described near-term business opportunities targeting customers as “sesame seeds,” a Chinese turn of phrase describing small gains not worth chasing at the expense of a bigger payoff.

    Editor: Marianne Gunnarsson.

    (Header image: VCG)