DeepSeek V4.1 Flash 1017, the version DeepSeek serves under its deepseek-flash API name, processed 61.2 trillion tokens on OpenRouter in the week through 16 November, the most of any model and 3.8 times the 16 trillion of second-placed Tencent Hy4, by our arithmetic on OpenRouter’s rankings. Over the latest day, DeepSeek’s own servers carried 12.4% of that model’s OpenRouter tokens, and 38 other providers carried the rest, according to the model’s OpenRouter page.
The same week, DeepSeek began rationing its free chatbot. Users told STAR Market Daily, a title of the Cailian Press group, on 16 November that the app and website cited “算力不足”, insufficient computing power in our translation, as the reason long conversations inactive for more than 7 days are no longer supported, and that a single input was cut to about 40,000 characters, as we reported on Monday.
Business Pill · Model Routing explained
An evergreen Business Engineer explainer on model routing, sending each request to the right AI model. It teaches the general idea only and does not discuss DeepSeek or OpenRouter.
The key insight: As we read it, DeepSeek’s compute limit sits in its own servers, not in demand for its model: V4.1 Flash 1017 led OpenRouter with 61.2 trillion tokens in a week, yet DeepSeek’s endpoint carried 12.4% of the model’s OpenRouter tokens over the latest day, and 38 other providers carried the rest.
What OpenRouter’s Numbers Show
The 61.2 trillion tokens were up 10% on the week before, OpenRouter’s rankings show. Over the latest complete day the model processed 9.13 trillion tokens, up 27% on the day before, and 201 trillion over the trailing 30 days. The earlier V4.1 Flash version processed another 15.5 trillion tokens in the week, down 12%.
By model author, DeepSeek had 31.8% of OpenRouter’s text requests in the week beginning 9 November, ahead of OpenAI at 19.0% and Google at 15.1%, the rankings page says.
Yicai reported on 16 November that V4.1 Flash 1017 processed 59.2 trillion tokens on OpenRouter in the week from 9 to 15 November, the most worldwide, against 15.8 trillion for V4.1 Flash in the week of 14 September. Yicai wrote, in our translation, that DeepSeek’s build-out of computing power had not kept pace with the demand curve.


What Was Posted on X
The X account @aleabitoreddit posted this at 00:22 UTC on 17 November, re-sharing STAR Market Daily’s 16 November report that DeepSeek is limiting use of its app and website, with screenshots of the Chinese item and an English rendering. The piece checks the report’s wording and adds OpenRouter’s own usage figures; the post’s view on what shortages signal is the poster’s, not ours.
Who Serves DeepSeek’s Model
OpenRouter’s API lists 50 endpoints from 39 providers for V4.1 Flash 1017, one of them DeepSeek. The model page explains: “Different companies host the same model”, and OpenRouter routes each request to one of them.
Over the latest day, DeepSeek’s endpoint had a 12.4% token share, the largest single share, followed by Together at 8.8% and Prime Intellect at 8.0%. The other 38 providers carried 87.5% between them, by our sum of the page’s figures, which add up to 99.9% because of rounding. DeepSeek’s listing charges $0.15 per million input tokens and $0.60 per million output tokens.
Other hosts have published their own work on the model: Baseten says its engine runs V4.1 Flash at about 2,060 tokens per second on eight B300 GPUs. Outside OpenRouter, DeepSeek reached 28.2% of open-weight tokens on Databricks in the week of 2 November, by our sums of the Databricks AI Index.
What the Limits Cover
Yicai reported that the limits apply to the consumer app and web versions, and that the API developers use has its own call limits and pay-per-use billing, which it described as a separate set of rules. It also reported that some users received the quota message after less than two hours of use.
When the user’s daily use hits the cap, the app shows “因服务器繁忙,使用额度已达上限”, or, in our translation, because the server is busy, the usage quota has been reached, STAR Market Daily reported. Yicai wrote that DeepSeek had not published unified rules for the limits or said when they would end, as of its publication.
DeepSeek’s status page read “Everything is running smoothly” when we checked at 00:35 UTC on 17 November, and listed uptime from August to November of 99.56% for its chat service and 99.55% for the V4.1 Flash API.
The Structural Read
As we read the endpoint table, publishing the weights lets other companies sell the capacity DeepSeek lacks: 39 providers list the model on OpenRouter, and the six largest endpoints, DeepSeek’s included, carried 46.1% of its tokens over the day, by our arithmetic.
As we read Yicai’s report, the rationing falls on the free app and website, while the API, which bills by use, keeps its own limits; the cap lands on the users who pay nothing.
OpenRouter, rankings page
“Models are ranked by tokens processed through the OpenRouter API by millions of developers.”
Three Implications
FOR DEEPSEEK Users told STAR Market Daily the app and website now cap long conversations, single inputs and daily use, and Yicai reported the API is under a separate set of rules.
FOR THE HOSTS OpenRouter lists 50 endpoints from 39 providers for V4.1 Flash 1017; Together carried 8.8% of its tokens over the latest day and Prime Intellect 8.0%.
FOR OPENROUTER USERS DeepSeek had 31.8% of OpenRouter’s text requests in the week beginning 9 November, ahead of OpenAI at 19.0% and Google at 15.1%, the rankings page says.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Open vs Closed Meta-Framework.
The framework says: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read the OpenRouter figures, DeepSeek has opened the model layer, which 38 other providers now host, while the computing power behind its own free chat is the layer in short supply; the framework predates these figures and does not discuss DeepSeek’s limits.
What Is Not Established
OpenRouter’s figures cover traffic routed through OpenRouter. They leave out requests sent straight to DeepSeek’s own API and use of its app, so the 12.4% is a share of one channel over one day, not of all V4.1 Flash usage.
The limits are known from users’ reports and the in-app messages they quoted to STAR Market Daily and Yicai. When we checked at 00:35 UTC on 17 November, the latest item on DeepSeek’s API news page was dated 13 November. We draw no link between the limits and the US remote-access rule released on 16 November.
The Bottom Line
DeepSeek V4.1 Flash 1017 led OpenRouter with 61.2 trillion tokens in the week through 16 November, while DeepSeek’s own servers carried 12.4% of the model’s OpenRouter tokens over the latest day. In the same week, users reported that DeepSeek’s free app and website had begun capping long conversations and daily use, citing insufficient computing power.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The usage figures come from OpenRouter’s rankings page, its page for DeepSeek V4.1 Flash 1017 and its public endpoints API, which we read at 00:41 to 00:42 UTC on 17 November; the 3.8 times ratio and the 87.5% share are our arithmetic on them. The rationing details come from STAR Market Daily’s 16 November report, as shown in an image attached to a post on X, and from Yicai’s 16 November article, which we read in full in Chinese; translations are ours. We also read DeepSeek’s status page and API news page. We did not contact DeepSeek or OpenRouter for this piece.
Sources: OpenRouter: LLM rankings (usage through 16 Nov 2026) · OpenRouter: DeepSeek V4.1 Flash 1017 model page · OpenRouter API: V4.1 Flash 1017 endpoints · Yicai (16 Nov 2026, Chinese) · DeepSeek status page









