Telegraph

7x24 real-time news feed, auto refreshBeijing Time：

All A·Important B·Normal #DeepSeek V4 ✕
B
[DeepSeek webpage and API services have been restored and some service performance is still degraded] On November 3, DeepSeek’s official website updated the status announcement stating that the previous partial interruption of webpage and API services has been repaired, and the repair effect is currently being monitored. However, there are still performance degradations in the API services and conversation services of DeepSeek V4 Pro and DeepSeek V4.1 Flash.
#DeepSeek V4#DeepSeek11-03 17:16
B
[A DeepSeek large model has the highest number of calls in a single week in the world] October 26th, according to the latest data from OpenRouter, the total number of global AI large model calls last week (October 19th to 25th) was 188 trillion Tokens, a month-on-month increase of 14.63%. Among the large AI models on the list, the weekly use volume of China's large AI models reached 93.54 trillion Tokens, a month-on-month increase of 18.14%; during the same period, the weekly use volume of large American AI models was 22.47 trillion Tokens, a month-on-month increase of 29.44%. The number of weekly calls for large models in China has surpassed that of the United States for 26 consecutive weeks, ranking first in the world. The reporter noticed that last week, the top five global calls were all large Chinese AI models. Among them, DeepSeek V4.1-Flash-1017 ranked first, with weekly calls reaching 33.6 trillion Tokens, a month-on-month increase of 801%. (Daily Economic News)
#DeepSeek V4#DeepSeek#large model10-26 13:54
B
[The latest global AI large model weekly token call list is released: Step 5 Preview ranks among the top three in the world] October 19th, according to the latest data from OpenRouter, the total number of global AI large model calls last week (October 12th to 18th) was 164 trillion Tokens, a month-on-month decrease of 4.65%. Among the large AI models on the list, the weekly use volume of China's large AI models reached 79.18 trillion Tokens, a month-on-month decrease of 8.72%; during the same period, the weekly use volume of large American AI models was 17.36 trillion Tokens, a month-on-month decrease of 8.49%. The number of weekly calls for large models in China has surpassed that of the United States for 25 consecutive weeks, ranking first in the world. The reporter noticed that last week, the top five global calls were all large Chinese AI models. Among them, DeepSeek V4.1 Flash ranked first for two consecutive weeks, with a weekly call volume of 32.8 trillion Tokens, a 20% month-on-month decrease; GLM 5.3 Flash ranked second, with a weekly call volume of 12.2 trillion Tokens; Step 5 Preview rose to third, with a weekly call volume of 11.5 trillion Tokens, a month-on-month increase of 30%. (Daily Economic News)
#DeepSeek V4#DeepSeek#Step stars#large model#GLM#Wisdom spectrum#AI10-19 10:31
B
[DeepSeek V4.1 Flash releases new version] "Science and Technology Innovation Board Daily" reported on the 17th that a reporter from "Science and Technology Innovation Board Daily" noticed that DeepSeek V4.1 Flash has been updated to version V4.1-Flash-1017, and the official web page, APP and API models have been synchronized to the new model. Compared with the previous version of DeepSeek V4.1 Flash, the new model has achieved improvements in all aspects of benchmark testing. The reporter learned that this upgrade focuses on optimizing platform frameworks such as DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. (Kechuangban Daily reporter Huang Xinyi)
#DeepSeek V4#DeepSeek#Tencent10-17 16:35
B
[DeepSeek web client and App have resumed service] "Science and Technology Innovation Board Daily" reported on the 15th that after testing by "Science and Technology Innovation Board Daily", DeepSeek web client and App have resumed service. The service interruption lasted nearly 5 hours. After 8 o'clock this morning, a large number of netizens reported that DeepSeek web and App experienced service abnormalities, and the page prompted "the server is temporarily unavailable." According to DeepSeek's official response, this service exception has affected the Chatservice, DeepSeek V4.1 Flash API Service (API Service) and one other component. (Kechuangban Daily reporter Huang Xinyi)
#DeepSeek V4#DeepSeek#server#AI10-15 13:12
B
[China Mobile Cloud Company releases heterogeneous hybrid reasoning system: brain-like chip + domestic GPU to break through the performance bottleneck of a single architecture] On September 13th, Du Yujian, technical director of the Brain-like and Light Laboratory of China Mobile Cloud Company, released a heterogeneous hybrid reasoning system for large model reasoning at the 2026 Computing Power China Exhibition. It adopts the "brain-like chip + domestic GPU" architecture and benchmarks NVIDIA's Vera Robin + Groq solution. This system uses a combination of Lingxi Technology's brain-inspired chips and Tianshu Intelligent Core GPU. In the DeepSeek V4 Flash inference scenario, the performance is more than doubled compared to similar GPU clusters, operating costs are reduced by more than 40%, and it is autonomously controllable. Du Yujian said that the introduction of brain-inspired computing power is equivalent to "turbocharging" existing GPU resources. (Reporter Guo Songqiao)
#DeepSeek V4#DeepSeek#NVIDIA#GPU#large model#chip#Computing power09-13 13:43
B
[DeepSeek V4.1 Flash model officially released] On September 10th, DeepSeek officially released the DeepSeek V4.1 Flash model. It is the smallest model in the company’s new model structure series, featuring native multi-modal visual understanding. According to reports, the new generation model significantly reduces the size of the KV Cache. Compared with the previous generation model, the demand for HBM is reduced to 1/4 and the demand for SSD is reduced to 1/8. In Agent usage scenarios, the cost of cache hits often accounts for a relatively high proportion, and the compression of KV Cache greatly reduces the usage cost of Agent tasks.
#DeepSeek V4#DeepSeek#Agent#multimodal#Futures Market Intelligence#HBM09-10 14:07
B
[New model will be launched soon. DeepSeek will merge fast, expert, and image recognition modes] "Science and Technology Innovation Board Daily" reported on the 10th. A reporter from "Science and Technology Innovation Board Daily" discovered that DeepSeek's interface has merged fast, expert, and image recognition modes, and there is no longer a separate expert mode. The DeepSeek open platform previously announced that DeepSeek plans to release the V4.1 Flash model around September 10, 2026, Beijing time, supporting native multi-modality. DeepSeek plans to offline the V4 Pro service at 12:00 on September 14, 2026, Beijing time. At that time, DeepSeek V4 Pro will be routed to V4.1 Flash and billed according to V4.1 Flash. (Kechuangban Daily reporter Huang Xinyi)
#DeepSeek V4#DeepSeek#multimodal#AI#Artificial intelligence subverts the future09-10 10:49
B
[DeepSeek V4.1 Flash starts internal testing of the intermediate version] "Science and Technology Innovation Board Daily" reported on the 8th, "Science and Technology Innovation Board Daily" learned that DeepSeek V4.1 Flash has started internal testing of the intermediate version. According to reports, it adopts a new model structure, native multi-modal support, stronger capabilities, faster speed, and lower cost. (Kechuangban Daily reporter Huang Xinyi)
#DeepSeek V4#DeepSeek#multimodal#AI09-08 15:47