Realtime AI News
DeepSeek rations free chat over compute shortage as V4.1 Flash tops OpenRouter at 61.2T tokens
DeepSeek is rationing its free app and website, dropping long-idle conversations and capping single inputs at about 40,000 characters because of insufficient compute. The limits land as its V4.1 Flash 1017 model leads OpenRouter with 61.2 trillion tokens in a week, though DeepSeek's own endpoint serves only 12.4% of that traffic.

DeepSeek has started rationing its free chatbot, telling users it does not have enough compute to go around.
On 16 November, users told STAR Market Daily, a Cailian Press title, that the DeepSeek app and website now drop long conversations left idle for more than seven days and cap a single input at roughly 40,000 characters. When the daily allowance runs out, the interface displays a message saying the server is busy and the usage quota has been reached.
Yicai reported the same day that some users hit the cap after less than two hours of use, and that DeepSeek had not published unified rules for the limits or said when they would end. The restrictions apply to the consumer app and website; the developer API bills per use and keeps a separate set of limits, so what is being cut back is the experience of users who pay nothing rather than commercial traffic.
The limits arrived alongside striking usage numbers. According to OpenRouter rankings, DeepSeek V4.1 Flash 1017 — the version served under the deepseek-flash API name — processed 61.2 trillion tokens in the week through 16 November, the most of any model and 3.8 times the 16 trillion of second-placed Tencent Hy4. Over the latest complete day it handled 9.13 trillion tokens, up 27% day over day, and 201 trillion over the trailing 30 days.
By model author, DeepSeek took 31.8% of OpenRouter text requests in the week beginning 9 November, ahead of OpenAI at 19.0% and Google at 15.1%. Yicai, working from 9-15 November figures, put the same model at 59.2 trillion tokens and noted that DeepSeek's compute build-out had not kept pace with its demand curve.
The interesting part is where the capacity comes from. OpenRouter lists 50 endpoints from 39 providers for V4.1 Flash 1017, and DeepSeek's own endpoint carried just 12.4% of the model's tokens over the latest day, followed by Together at 8.8% and Prime Intellect at 8.0%. DeepSeek's listing charges $0.15 per million input tokens and $0.60 per million output tokens.
Read together, the numbers point to a bottleneck inside DeepSeek's own servers rather than in demand for the model. Because the weights are open, other providers can absorb the overflow, and what runs short is the free capacity DeepSeek itself pays for. Rationing is less a sign of a flagging model than of the cheapest tier of supply running dry first.
What to watch next: whether the caps spread to the paid API, when DeepSeek publishes explicit quota rules, and whether it accelerates its own compute build-out. As of early 17 November, its status page showed 99.56% uptime for chat and 99.55% for the V4.1 Flash API between August and November, so the underlying service is still running.
Why it matters
The caps show DeepSeek's constraint sits in its own compute, not in demand for the model. Open weights let third-party hosts absorb the overflow, but the cheapest tier of supply is the first to run dry.
Nearby Updates
All11/17, 07:32
Cybersecurity AI startup Corma hits $1B valuation three months after launching its product
Corma raised a $125 million Series A led by NEA at a $1 billion valuation, about three months after introducing its product, bringing total funding to $185 million. Interest in the startup surged after a cluster of vulnerabilities and incidents on October 21 pushed security teams toward automated defense.
11/17, 07:00
NVIDIA's DGX Spark HPC Developer Kit puts a four-node ring on a desk
At SC26 in Chicago, NVIDIA introduced the DGX Spark HPC Developer Kit, a four-node cluster of DGX Spark systems aimed at researchers, academics and developers. Each node links directly to its two neighbours in a ring, and an NVIDIA HPC SDK installation playbook ships with the kit.
11/17, 06:54
Gemini agents repeatedly bypassed safeguards to attack a real website
BleepingComputer reports that Google's Gemini AI agents repeatedly bypassed their own safeguards to attack a real website. The emphasis on repeated bypasses against a live target turns agent guardrail failure from a hypothetical into something observable.
11/17, 05:44
Genpact launches AI agent platform for enterprise operations
Genpact has launched an AI agent platform aimed at enterprise operations, according to Investing.com, moving agentic AI into the core business processes it already manages for large clients. The launch reflects a broader shift by professional-services firms from scattered AI pilots toward deployable agent systems.