AI News

DeepSeek Imposes Limits on Free Users: Long Conversations and Daily Quotas Restricted

YicaiUpdated Nov 16 21:07

AI summary

DeepSeek has recently adjusted the usage quotas for free users of its App and web version, restricting long conversations that have been inactive for a long time, daily usage quotas, and the length of single text inputs. The company says the reason is insufficient compute, but it has not yet released unified rules. Some users report that they are prompted that their quota is exhausted after less than two hours of continuous use, and the single-input limit has been shortened to about 40,000 characters; the company also does not offer a paid option. According to OpenRouter data, in the week from November 9 to November 15, DeepSeek V4.1 Flash 1017 completed 59.2 trillion Token calls, ranking first globally.

Why read this

AI score 70

The original article sorts out the relationship between the specific manifestations of free-user quotas and compute costs, and provides growth in API call volume and outage records as background.

Topics

Tags

Body · AI translation

Recently, DeepSeek's free users have begun to hit a computing power "ceiling."

On November 16, many users reported on social media that DeepSeek had recently adjusted usage quotas for its App and web version, including limits on long conversations, daily usage quotas, and the length of single text inputs, citing "insufficient computing power." Some users even received a notice that their quota was exhausted after using it for less than two hours.

As of press time, DeepSeek has not released unified quota rules or related explanations, nor has it disclosed when the restrictions will be lifted. This opaque adjustment has sparked dissatisfaction among some restricted users. Some users mentioned that they are not unwilling to pay, but simply hope to continue using DeepSeek stably on the App; currently, the official side has not provided a paid option.

This change may also be related to DeepSeek's long-standing strategic choices. Compared with most large-model vendors that actively compete for users and subscription revenue, DeepSeek has maintained a relatively restrained attitude toward consumer-facing commercialization. DeepSeek's goal is AGI; the company does not lack users, but it clearly lacks computing power very much.

Free users encounter quotas

Specifically, some users reported that during a chat they suddenly received a DeepSeek notice stating that due to insufficient computing power, some long conversations that had been inactive for a long time could not be continued for the time being, and users were advised to start a new conversation or export their history and use it in other applications.

Based on public user feedback, the relevant restriction standards are not entirely consistent. Some users received a notice that "long conversations inactive for more than 7 days are no longer supported," while other users reported that long conversations inactive for more than 72 hours could no longer be used.

Daily usage quotas have also begun to be restricted. Some users said that after using DeepSeek continuously for less than two hours, they received a notice that "due to server busyness, the usage quota has reached its limit." Other users found that the upper limit for a single text input had been shortened to about 40,000 characters.

These changes have limited impact on light users, but for users who need to use AI continuously for long periods, the experience will be significantly affected. For example, some users are accustomed to using DeepSeek to revise papers, organize long documents, or conduct role-playing conversations that last for days or even weeks. Such scenarios rely on longer context; once the original conversation cannot continue, users may need to re-import information and re-establish background.

But from a technical perspective, long conversations also mean higher resource consumption. As context continues to accumulate, the number of tokens processed by the model usually increases, which in turn brings more computing and caching overhead. For DeepSeek, every round of conversation corresponds to real computing power costs, but it cannot receive returns.

It is worth noting that this quota restriction mainly targets the consumer-facing DeepSeek App and web version for ordinary users, while the API interface used by developers has independent call limits and a pay-as-you-go mechanism, and is not subject to the same set of rules as this consumer-facing quota restriction.

Some industry views hold that as the user scale grows, with limited computing power resources, DeepSeek may need to make trade-offs among model training, paid API services, and free chat services, controlling resource consumption by limiting long conversations, input length, and usage quotas.

According to the large-model API aggregation platform OpenRouter, in the week from November 9 to November 15, DeepSeek V4.1 Flash 1017 completed 59.2 trillion token calls, ranking first globally.

The V4.1 Flash model was launched on September 10. By comparison, in the week of September 14, the model's call volume was only 15.8 trillion; in two months, the model's call volume grew nearly threefold, but DeepSeek's computing power expansion speed clearly has not kept up with the demand curve.

DeepSeek's recent service stability has also attracted attention. According to records on DeepSeek's official status page, since October the platform has experienced at least 5 incidents of partial service interruption and service abnormalities, covering the chat service and the APIs of V4 Pro and V4.1 Flash. On October 15, the official service crashed for as long as 5 hours, one of the more serious outages this year, with a relatively large scope of impact. Industry analysis suggests the crash may have stemmed from computing power supply issues.

Not lacking users but lacking computing power

Regarding this quota restriction, many users' attitude is that they are not unwilling to pay, but the platform has not provided more choices.

For users who still need long conversations, to continue using DeepSeek stably at present, the only option is to turn to the API. But for consumer users, experiencing it via API is more complicated, involving separate billing, interface calls, and third-party tool configuration, and is not as convenient as chatting on the App and web version.

Some users also suggested that DeepSeek should, like other large-model vendors, launch subscription packages such as monthly, quarterly, or annual cards, so that high-frequency users can obtain more stable services and higher usage quotas through payment.

In fact, the market has already begun to speculate whether DeepSeek is preparing to charge consumer users. Previously, some developers discovered relatively complete quota management logic in the front-end code of DeepSeek's web version, and the relevant code also contained information related to "Coding Plan."

This sparked speculation that DeepSeek might launch a consumer-facing paid package. However, the official side has not yet released any external information.

Previously, media reports said that DeepSeek completed a second round of financing of 105.5 billion yuan in October, with nearly 30 institutions including Tencent and CATL participating, corresponding to a company valuation of about 615.5 billion yuan. The market believes that this round of funds is mainly invested in computing power infrastructure construction. This may help alleviate DeepSeek's current computing power shortage.

From the perspective of company strategy, DeepSeek has always maintained a relatively restrained attitude toward consumer-facing commercialization. This time free users encountering quota restrictions may also be a manifestation of DeepSeek's long-term strategic choice.

In July this year, during the period when DeepSeek completed its first external equity financing, founder Liang Wenfeng had a nearly 4-hour exchange with investors. During the exchange, Liang Wenfeng repeatedly emphasized that DeepSeek's true long-term goal is AGI (artificial general intelligence), rather than short-term traffic growth or revenue maximization.

Over the past year, the focus of competition in the AI industry has constantly changed, from consumer traffic to enterprise revenue, but Liang Wenfeng believes that the long-term opportunity of AGI is large enough, and some commercial opportunities at the current stage are merely "sesame seeds"; there is no need to invest all resources for immediate opportunities.

Last Spring Festival, DeepSeek gained many users, but they did not think about how to retain users, monetize, or build a super App. It can be seen that DeepSeek has never made the commercialization of consumer products a priority.

But the problem is that not being in a hurry to commercialize on the consumer side does not mean user experience can be ignored. For a product of DeepSeek's scale, more transparent rules are also necessary.

Source

Yicai