Inter-office memo

To
Everyone who uses DeepSeek’s app or website
From
Mo, Chief Executive
Date
November 17, 2026, 9:09 AM CST
Re
DeepSeek’s chat limits

DeepSeek’s free chat now has a daily limit, users report. What changed, and what to do

Since November 15, 2026, users of DeepSeek’s free app and website report a daily quota, messages capped at about 40,000 Chinese characters, and long chats idle over seven days that can’t be continued, “because computing power is insufficient.” DeepSeek hasn’t announced the limits or said when they end. No report covers its pay-per-token API, which DeepSeek’s open-source chat demo uses to carry on your chats.

What changed

On Sunday, November 15, 2026, people using DeepSeek’s free chat, in its app and on its website, began reporting three new limits. IT Home collected them the next morning, and TechNode, 36Kr and Pandaily followed. In IT Home’s account, translated:

  • A daily quota. When it runs out, the page says the usage quota has reached its limit “because the servers are busy.”
  • Shorter messages. A single message is capped at roughly 40,000 Chinese characters. None of the reports says how that counts English or code.
  • Week-old chats close. A notice says that “because computing power is insufficient,” long conversations inactive for more than seven days are no longer supported. 36Kr reports they can’t be resumed.

The limits cover the consumer chat. None of the reports we read says the API changed.

What DeepSeek has and hasn’t said

Nothing in public. When IT Home published, DeepSeek had made no announcement and given no date for lifting the limits, and Pandaily put it plainly: “DeepSeek has not issued an announcement or confirmed any of the limits, and the figures below come from users, not the company.” By November 17 we found no statement either. DeepSeek’s changelog says nothing about chat quotas, and its status page says “All systems are operating as expected.”

DeepSeek hasn’t said:

  • how large the daily quota is, or when it resets;
  • what makes a conversation “long,” or whether a closed one can still be read;
  • whether the limits are permanent, or apply to every user and country.

The reports follow weeks of heavy demand. IT Home counts eight incidents on DeepSeek’s status page since October, affecting chat and the V4 Pro and V4.1 Flash APIs, and says V4.1 Flash handled 33.6 trillion tokens on OpenRouter in the week of October 19 to 25, up 801% on the week before. Industry analysts quoted by 36Kr think closing idle chats frees the memory they hold for people chatting now. That’s their reading, not DeepSeek’s.

Is a paid plan coming?

Perhaps, but nothing is announced. On November 14, IT Home reported that developers had found quota code in DeepSeek’s web chat: a banner for when a quota runs out, a check of each model’s remaining quota every five minutes, and a package named “Coding Plan.” As Pandaily sums it up, “DeepSeek has announced no subscription, price, tier or launch date,” and IT Home called a first paid plan for individuals community speculation. Code in a web app can also belong to features that never ship.

For now, the only way for individuals to pay DeepSeek is its API, billed by the token from a balance you top up.

What the API costs

DeepSeek’s pricing page, per million tokens, on November 17, 2026:

deepseek-flash (V4.1 Flash-1017) deepseek-v4-pro (V4 Pro-0813)
Input the API has seen before (cache hit) $0.006 peak, $0.003 off-peak $0.044 peak, $0.022 off-peak
New input (cache miss) $0.30 peak, $0.15 off-peak $1.32 peak, $0.66 off-peak
Output $1.20 peak, $0.60 off-peak $3.96 peak, $1.98 off-peak
Context 1M tokens 1M tokens
Requests at a time, per account 2,500 500

Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, which is 9 AM to noon and 2 to 6 PM in Beijing. Weekends and Chinese public holidays are off-peak all day. Web searches and page fetches cost $0.75 per 1,000, and reading files costs $0.225 per 1,000 pages.

One thing to know before moving a long chat to the API: DeepSeek’s chat API is “stateless,” so each message resends “all previous conversation history.” Context caching bills the part it has already seen at the cache-hit rate. By those prices, resending a 200,000-token conversation on Flash costs 6 cents at the peak cache-miss rate and about a tenth of a cent on a cache hit.

What to do

  1. Keep long work in a chat you’re using. The notice covers conversations left inactive for more than seven days. For anything older, paste a short summary into a new chat rather than counting on the old one.
  2. Export what matters now. DeepSeek’s chatbot guide refers to “the chat history you exported from DeepSeek Chat,” but doesn’t document the steps, so look in DeepSeek Chat’s settings.
  3. Carry on through the API with DeepSeek’s demo. On November 13, DeepSeek opened its app’s search, file reading, link reading and voice features to developers and published an open-source chat demo. With it, DeepSeek says, you can “import the chat history you exported from DeepSeek Chat and continue your conversations.” Open the demo, enter an API key from the DeepSeek Platform, and import the ZIP. Know its limits: “an import replaces the existing history,” your key sits in the browser’s local storage, and DeepSeek calls it “for demonstration only.” By its import notes, an attachment without its original file keeps only its name and size, and the model doesn’t see it again.
  4. For code, use the API or a plan made for agents. DeepSeek’s Claude Code guide points Claude Code at https://api.deepseek.com/anthropic with the model deepseek-flash[1m]. OpenCode Go, $10 a month, includes DeepSeek V4.1 Flash 1017 and V4 Pro, with an estimated 43,300 Flash requests every five hours. Enjoy runs OpenCode beside Claude Code, Codex and the other labs’ agents with every model from the providers you set up in it, and lets you switch agents mid-conversation when one runs low.
  5. Run a model yourself. DeepSeek’s weights are free under the MIT license, but V4.1 Flash-1017 has 763 billion parameters, far beyond a laptop’s memory. Smaller open models do run on one, with no quota at all; our local coding agent guide compares the ways.

We’ll update this memo when DeepSeek says how the limits work.

6.Questions.

Does DeepSeek have a daily limit now?

Yes, by users’ reports since November 15, 2026. When the free app or website reaches its daily quota, it says the usage quota has reached its limit because the servers are busy. DeepSeek hasn’t announced the quota, how large it is, or when it resets.

Why can’t I continue an old DeepSeek conversation?

Because of a new rule users report. A notice says that because computing power is insufficient, DeepSeek no longer supports long conversations inactive for more than seven days. Start a new chat, or export your history and carry on through DeepSeek’s API with its open-source chat demo.

Is there a paid DeepSeek plan with more usage?

Not yet. Developers found code in DeepSeek’s web chat that mentions a “Coding Plan,” IT Home reported on November 14, but DeepSeek has announced no subscription, price or date. Today the only way for individuals to pay DeepSeek is its API, by the token.

Is the DeepSeek API limited too?

Not in any report we found. The API has its own limits, 2,500 requests at a time per account on deepseek-flash and 500 on deepseek-v4-pro, and on November 17 DeepSeek’s status page said “All systems are operating as expected,” chat included.

How much does DeepSeek’s API cost?

deepseek-flash costs $0.30 per million input tokens and $1.20 per million output tokens at peak, and half that off-peak. Input the API has already seen costs $0.006 per million at peak. Web searches are $0.75 per 1,000.

Can I run DeepSeek on my own computer instead?

Not the model the app uses, on most computers. DeepSeek V4.1 Flash-1017’s weights are free under the MIT license, but it has 763 billion parameters, which takes server hardware. Smaller open models do run on a laptop; our guide compares the ways.

Enjoy Agentic Industries · Checked against its sources on