DeepSeek length limit reached? The chat hit a 1M token wall, not a daily cap, and a handoff gets you past it

How do you fix "Length limit reached" on DeepSeek?

A full DeepSeek chat can often still do one last job: write its own handoff. DeepSeek has no button that extends a chat, so the way on is a new chat that carries the work across. It takes about a minute:

1. In the old chat, edit one of your own recent messages and replace it with the handoff prompt below. An edit cuts the thread back to that message, so the request may fit again. DeepSeek doesn't document this, and users have reported edits capped at about six since 29 May 2026, so use one near the end.
2. If the edit is refused too, copy what you need by hand: the latest draft or code, your rules, the decisions. Paste them into a new chat with the same prompt.
3. Copy the handoff DeepSeek writes.
4. Open a new chat in the same mode (Expert or Instant), paste the opening line below with the handoff under it, and read its three line reply before you ask for anything new.

```text

This conversation is moving to a new chat that cannot see it. Write a handoff I can paste in as the first message, under 800 words, with these parts:

1. Goal: what we are making or deciding, in two sentences.
2. Latest version: the current draft or code, quoted in full if it is under 300 words, otherwise its outline with every heading or function name.
3. Decisions: one line each, with the reason.
4. Ruled out: what we tried or rejected, and why.
5. My rules: tone, format, length, language, and any names or numbers I asked for.
6. Facts to keep exactly: figures, dates, file names, links.
7. The next step and any open questions.

Add nothing we did not discuss. Mark anything you are unsure of with [check].

```

```text

Below is a handoff from a DeepSeek chat that hit its length limit. Read it, then reply in three lines only: the goal, the rules you will follow, and the next step. Wait for my go before doing anything else.

[paste the handoff here]

```

Two parts of that prompt do most of the work. The latest version is quoted, not summarised, because a summary of your draft is a different draft. And the ruled out list travels with it. A new chat has no record that an idea already failed. What doesn't work is pasting the whole old transcript. A V2EX user whose DeepSeek web chat ended after 243 rounds in October 2026 tried to feed a new chat the old one's share link. The page loaded lazily, so the new chat never read it all, and he expected the full history to overflow again anyway. He was right to. A complete transcript is exactly as big as the chat that just filled up.

How long can a DeepSeek chat be before the length limit?

A DeepSeek chat can hold 1M tokens as of October 2026. DeepSeek's V4 launch notes of 24 April 2026 put it in one line: "1M context is now the default across all official DeepSeek services." Its API pricing page, read on 18 October 2026, lists a 1M context length for both current models. Expert mode in the app runs DeepSeek V4 Pro, whose 0813 release reached app and web on 13 August 2026, and DeepSeek V4.1 Flash 1017 replaced the app's Flash model on 17 October. ChatGPT Free's Instant window is 27K tokens on OpenAI's pricing table (as archived on 13 October 2026). A free DeepSeek chat holds as much as Claude's paid plans.

In characters, that's a lot of typing. DeepSeek's token usage page says "1 English character ≈ 0.3 token" and "1 Chinese character ≈ 0.6 token". On that math a full chat holds about 3.3 million English characters, or about 1.7 million Chinese characters. DeepSeek adds that the ratio varies by model, so read both as estimates. Few people type that much into one thread. The wall usually comes from what gets pasted or attached, which the next sections cover.

The window also grew fast, which is why guides disagree on the number. Each row below is the window listed for that model at launch, by the source named in the row:

So a page that quotes 64K is describing R1, and one that quotes 128K is describing the V3 era. Both were right once. Neither is right for a chat you started after April 2026.

Is the DeepSeek length limit a daily limit?

No. The DeepSeek length limit counts the size of one chat, not how much you've used today, so a new chat works at once and waiting does nothing for the old one. DeepSeek's app shows three different walls, and only one is about length. The length message's English wording below is as quoted in a Better DeepSeek issue title of 22 June 2026, and its Chinese as quoted in a DeepSeek-V3 bug report of 12 June 2026. The busy message's Chinese is as Phoenix Tech quoted it during the 8 May 2026 outage.

DeepSeek hasn't published the regenerate and edit caps. The figures are user reports that Huxiu and TMTPost gathered on 30 May 2026, the day after people first hit them. Both quote 白墙, a Xiaohongshu account they call semi official, which advised a 15 to 30 minute wait. It warned that rapid clicking can make the block last longer. The busy message is a different animal, DeepSeek's HTTP 503 for high traffic, and DeepSeek server busy covers its peak hours.

DeepSeek hasn't explained the length message either. Nobody from the company answered the June bug report, which complained that the notice arrives with no warning. A bot closed it as stale on 11 August 2026.

Why does a DeepSeek chat hit the length limit so soon?

A DeepSeek chat fills up because every turn carries the whole thread again. DeepSeek's multi round chat guide calls its chat API "stateless" and says "the user must concatenate all previous conversation history and pass it to the chat API with each request." A chat app built on that API does the same work for you. So every reply you get becomes part of what the next turn must carry.

Pasted documents and code fill it fastest. An attached file isn't free room, because the model can only answer from text it has read into the same window. Asking for a full rewrite of a long draft, again and again, stacks one complete copy per reply. Users report the wall at very different points. One user of the Better DeepSeek extension hit it about an hour into a chat in June 2026. The V2EX user above got 243 rounds.

Thinking costs less room than several ranking guides claim. DeepSeek's thinking mode guide says that in a request without tools, "the CoT content from previous turns will not be concatenated into the context in the next turn." With tools on, it says the reasoning from every earlier turn must be passed back. Search is a tool. DeepSeek doesn't say which rule its app follows when search is on, so a long chat with search may fill faster than a plain one.

The maintainer of Better DeepSeek put the limit plainly on GitHub in June 2026: "Like all LLMs, DeepSeek has a context window limit, and it's not something we can solve on the extension side." DeepSeek shows no meter for how full a chat is. The habit that works is one chat per task, and a handoff once a thread has run for hours. Don't wait for the wall.

Can you bypass the DeepSeek length limit with a bigger window?

No. Neither DeepSeek's API nor any host on OpenRouter sells a bigger DeepSeek window than the free app already has. On 18 October 2026, OpenRouter's model list carried 16 DeepSeek models served through 162 endpoints by 41 providers, DeepSeek itself among them. Not one endpoint went above 1,048,576 tokens of context, and several sold less. The DeepSeek API has the same 1M. So the API buys control, not room.

Control still matters. On the API you send the history yourself, so you can drop old turns, keep a running summary at the top, and never pay to carry one search result twice. One reply can run to 384K tokens of output, per DeepSeek's pricing page. As of 18 October 2026, V4.1 Flash costs $0.15 per million input tokens off peak and $0.30 at peak, and $0.60 or $1.20 per million output tokens. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays.

Other apps offer the same size with different behaviour at the edge. Claude's help center lists 1M tokens for its newest models on paid plans, and Google's Gemini limits page lists 1 million on Google AI Pro and Ultra, both read on 18 October 2026. Claude handles the edge for you: when a chat nears its limit, "Claude summarizes earlier messages to continue the conversation seamlessly", as long as code execution is on. That's the handoff above, done automatically. Claude conversation too long covers what happens there, and the context window comparison lists windows model by model.

Whizi, which publishes this page, takes the opposite trade. It shows no length limit message at all. Every model, DeepSeek V4.1 Flash on Pro included, gets a budget of 40,000 input tokens per turn, and a chat that outgrows it is trimmed to fit silently, with no notice. That's far less room than DeepSeek's 1M, so it loses for one huge document and suits a long run of shorter tasks. Pro is $29.99 a month for 2,500 credits, and V4.1 Flash costs 1 credit a message (every DeepSeek row is 1 or 2, per using DeepSeek in Whizi). If where your chats are stored matters more than room, is DeepSeek safe compares the two privacy policies.