DeepSeek Rations Free Chat, Caps Input at 40,000 Characters

DeepSeek has started rationing its free chatbot, cutting the text a user can paste into a single message to about 40,000 Chinese characters and setting a daily usage quota, according to users cited by China’s STAR Market Daily on 16 November. Singapore’s Lianhe Zaobao and Hong Kong’s Sing Tao carried the report the same day. Users said the app and website told them the cause was “insufficient computing power”.

The limits follow a run of service problems. DeepSeek’s own status page lists 10 incidents between 1 October and 16 November, lasting 637 minutes in all, by our arithmetic from the start and resolution times it publishes. DeepSeek had not posted a notice about the new limits on its status page or API news page when we checked at about 05:15 UTC.

Business Pill · GPUs Explained

An evergreen Business Engineer explainer on why AI runs on GPUs. It teaches the general idea only and does not discuss DeepSeek.

The key insight: As we read it, DeepSeek is rationing the free side of its business while its paid side already prices busy hours: users report a cap of about 40,000 characters per input and a daily quota, and its API charges twice as much in peak hours. Its status page lists 10 incidents and 637 minutes of disruption since 1 October, by our arithmetic, so the stated shortage has a record behind it.

What Users Say Changed

STAR Market Daily, a title of the Cailian Press group, reported that DeepSeek has limited long conversations in its app and web versions. Several users said DeepSeek displayed the message that, because of “insufficient computing power, long conversations inactive for more than 7 days are no longer supported”, according to the report.

The same users said the cap on a single text input was cut to about 40,000 characters. Lianhe Zaobao added that other users reported conversations inactive for more than 72 hours were also unsupported, a shorter window than the seven days in the first message.

Users also told the outlets that DeepSeek now limits daily conversation volume. When a user hits that limit, the app shows a message that reads, in our translation, “because the server is busy, the usage quota has reached its limit”. None of the reports gives the size of the daily quota.

Every incident listed on DeepSeek's own status page from 1 October to 16 November 2026, by duration from the l
Every incident listed on DeepSeek’s own status page from 1 October to 16 November 2026, by duration from the listed start to resolution, read at about 05:15 UTC on 16 November. The 637-minute total is our arithmetic.

Ten Incidents in Seven Weeks

DeepSeek’s status page lists seven incidents in October and three so far in November, by our count. Eight of the 10 name the web or chat service. The longest, a partial outage of the web and API on 15 October, lasted 292 minutes, about 4.9 hours; Lianhe Zaobao described it as an interruption of nearly five hours.

Two November incidents ran 110 minutes on 3 November and 45 minutes on 7 November, both listed as partial outages of the web and API. Lianhe Zaobao reported that during the 7 November outage the chat screen showed a “server busy, please try again later” message. The most recent entry is a 22-minute slowdown of the web and chat service on 14 November.

Lianhe Zaobao noted the difference this time: rather than an outage, DeepSeek itself cited a lack of computing power as the reason to limit use. The paper wrote that this has raised doubts about DeepSeek’s compute supply and about whether its free service for ordinary users can be sustained.

Grouped bar chart: DeepSeek API output price per million tokens, deepseek-flash $0.60 off-peak and $1.20 peak, deepseek-v4-pro $1.98 off-peak and $3.96 peak
DeepSeek API output list prices per million tokens, off-peak and peak, from DeepSeek’s Models & Pricing page, read on 16 November 2026. DeepSeek states that off-peak rates are half of peak rates.

The API Already Prices Scarcity

DeepSeek’s paying developers already face a price signal for busy hours. Its pricing page lists deepseek-v4-pro output at $3.96 per million tokens in peak hours and $1.98 off-peak, and deepseek-flash at $1.20 and $0.60. The page says off-peak rates are half of peak rates; peak hours run 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, about 21% of the hours in a week, by our arithmetic.

The API also carries per-account concurrency limits of 2,500 requests for deepseek-flash and 500 for deepseek-v4-pro, according to DeepSeek’s rate-limit page. Requests above the limit receive an HTTP 429 error. The page says customers can request more capacity at no additional cost.

Against Alibaba’s new flagship, DeepSeek’s V4 Pro is still cheaper at its peak price. Alibaba lists Qwen4-Max output at $4.50 per million tokens, as we reported this morning, so V4 Pro’s peak output price is 12% lower and its off-peak price 56% lower, by our arithmetic.

The Compute Behind It

On the demand side, DeepSeek’s share of OpenRouter top-10 tokens rose to 56%, we reported on 8 November, and on 13 November DeepSeek opened its app’s tools, including web search, file input and voice, to developers through its API.

Supply is the constraint the reports point to. On 15 November we covered a report that tied DeepSeek and three other labs to more than $500 million of Nvidia Blackwell chips, and on 4 November a report that DeepSeek V5 would shift pretraining to Huawei’s Ascend chips.

The Structural Read

As we read the status page, October was the heavier month: 460 minutes across seven incidents, against 177 minutes across three so far in November, by our arithmetic. The new limits arrive after the incidents, not during one; no incident was listed for 16 November when we checked.

As we read the pricing page, the peak window covers 35 of the 168 hours in a week, about 21%, by our arithmetic. Developers who can shift work outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays pay half.

DeepSeek API Docs, Models & Pricing, read 16 November 2026

“Off-peak rates are half of the peak rates.”

Three Implications

FOR FREE USERS Users report a cap of about 40,000 characters per input and a daily quota; the size of the quota has not been published, according to the reports.

FOR DEVELOPERS DeepSeek’s rate-limit page lists concurrency limits of 2,500 for deepseek-flash and 500 for deepseek-v4-pro per account, with more capacity available on request at no additional cost.

RELATED Where DeepSeek’s money comes from is in our piece on state investors taking 27% of its nearly $16 billion round.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of The Inference Economy.

The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”

As we read it, DeepSeek’s free chat is the case where that marginal cost meets no revenue: the reported quota and input cap limit queries on the free app, while the API charges $3.96 per million V4 Pro output tokens at peak. The framework predates these limits and does not discuss DeepSeek.

What Is Not Established

DeepSeek has not published a notice about the limits that we could find, so the 40,000-character cap, the seven-day and 72-hour windows and the daily quota rest on user reports carried by STAR Market Daily, Lianhe Zaobao and Sing Tao. We did not see the in-app messages ourselves. We do not know the input cap that applied before, the size of the daily quota, or whether the limits are temporary.

The 40,000 figure counts Chinese characters, while DeepSeek’s API context of 1 million is counted in tokens, so the two are not directly comparable. We did not contact DeepSeek for this piece.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The State of The Inference Economy

The Business Engineer analysis of inference as a revenue engine with a marginal cost on every query.

Read the Map of AI →

The Bottom Line

Users report that DeepSeek has capped single chat inputs at about 40,000 characters and set a daily quota, citing insufficient computing power. Its status page lists 10 incidents and 637 minutes of disruption since 1 October, by our arithmetic, while its API already charges twice as much in peak hours.

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. The usage limits rest on user reports carried by STAR Market Daily, Lianhe Zaobao and Sing Tao on 16 November 2026, which we read in Chinese; English renderings of the in-app messages are our translation. DeepSeek had posted no notice on its status page or API news page when we checked at about 05:15 UTC. Incident dates, titles and durations come from DeepSeek’s own status page, and API prices and limits from its own documentation, read the same day. Incident totals, shares and price gaps are our arithmetic. We did not contact DeepSeek.

Sources: Lianhe Zaobao, DeepSeek usage limits over computing shortage raise doubts about free model, 16 November 2026 · STAR Market Daily (Cailian Press), via Sina, 16 November 2026 · Sing Tao Headline, 16 November 2026 · DeepSeek status page, history (read 16 November 2026) · DeepSeek API Docs, Models & Pricing (read 16 November 2026) · DeepSeek API Docs, Rate Limit & Isolation (read 16 November 2026) · DeepSeek, Full-Suite App API Release, 13 November 2026

Scroll to Top
FourWeekMBA