All news

CChinaTalk00:35

AI score10

WarTalk: Kofman on Starshield rate limiting and the lessons Ukraine offers the US military

On the WarTalk podcast, Michael Kofman pointed out that SpaceX has imposed rate limiting on Ukraine's Starshield network, capping drone connection speeds at roughly 160 km/h and making them unstable, which has caused heavy losses of Ukrainian light and heavy medium-range strike drones operating at depths of 50–300 kilometers and has hindered the mass production of its jet-powered interceptor drones. He believes that the United States still holds leverage over the war, that Europe cannot replace the United States' role in NATO, and that procuring drones is meaningless without accompanying drone platoons and companies.

#Industry news#English text is being prepared.

EEast Money00:35

AI score38

Peng Wensheng: Industrial capital dominates AI innovation financing, US private credit plays an important role

Peng Wensheng pointed out that the dominant players in financing this round of AI large-model innovation have shifted from independent venture capital to industrial capital (CVC), with large tech companies leveraging their first-mover advantages in computing power and cash flow to bear the vast majority of AI capital expenditure. US private credit has provided important debt financing for data center construction and small and medium-sized tech startups, while China has a structural mismatch of debt in the guise of equity. He also noted that angel and seed-stage projects account for 41% of all venture capital deals in the US, compared with only 9% in China.

#Industry news#English text is being prepared.

PPhoenix Technology00:33

AI score48

China Mobile's 11.6 billion yuan domestic computing power centralized procurement: Dahua's subsidiary Huaqi Zhihui wins 2.63 billion yuan bid

Dahua's subsidiary Huaqi Zhihui won the bid for China Mobile's 2026-2027 next-generation AI super-node equipment centralized procurement, with a bid amount of approximately 2.63 billion yuan, corresponding to a 4,096-card share. The project procures a total of 20,480 AI super-node cards as well as 448 RoCE switches and 82 storage systems, with a total amount estimated at 11.6 billion to 11.7 billion yuan; Huakun Zhenyu, Baode Computer, Baixin Information, and Shenzhou Kuntai were also shortlisted. Huaqi Zhihui was established in 2022, computing power is a new business line of Dahua, and China Mobile is a shareholder holding more than 5% of Dahua's shares, so this transaction constitutes a related-party transaction.

Same story · 2 reports

#Industry news#English text is being prepared.

CcnBeta00:28

AI score73

Anthropic Files for IPO, Q3 Revenue Reaches 12 Times Year-Earlier Level

Anthropic's filing with the U.S. Securities and Exchange Commission shows that revenue for the third quarter ended September 30 was 12 times the year-earlier level, with a net loss of $23.3 billion, compared with $12.1 billion a year earlier. Operating expenses for the quarter rose to $17.5 billion, up 292% year over year; the company currently has $895 billion in committed spending, the vast majority related to compute and infrastructure. Anthropic filed for its IPO on October 26, and the IPO is expected to take place before the U.S. Thanksgiving holiday.

Same story · 128 reports

#Industry news#Anthropic

LLeikeji00:27

AI score35

Li Jiaqi livestream sells Qwen Office: 100,000 monthly membership cards sold in half an hour, about 2 trillion Tokens

Li Jiaqi's livestream began selling Qwen Office personal membership monthly cards, selling 100,000 in half an hour, totaling about 2 trillion Tokens. Qwen Office CEO Chen Yusen appeared in the livestream and gave live demonstrations of features including importing homework test papers to organize knowledge points, generating training mini-games, and turning livestream replays and bullet comments into business analysis reports with one click. On September 3 this year, Tmall already launched a Token recharge center.

Same story · 10 reports

#Industry news#Agents

LLeitech00:27

AI score39

Alipay's AHA Agent Multi-Device Interconnection Solution Wins World Internet Conference "New Shine" Scenario Award

Alipay's AHA (Agent Hub Access) agent multi-device interconnection solution won the "New Shine" Scenario Award at the 2026 World Internet Conference Wuzhen Summit. The solution builds a three-tier interconnection protocol system based on the open-source A2A, MCP, and A2UI protocols, supporting cross-brand, cross-device agent collaboration on tasks. It has already extended Abao's service capabilities to 6 types of devices and implemented over a thousand service scenarios. Alipay has joined with more than 20 companies including Qwen, Huawei, OPPO, BYD, and NIO to build the ecosystem, achieving cross-device collaboration with 5 major smartphone manufacturers, whose combined market share exceeds 70%.

Same story · 2 reports

#Industry news#Agents#English text is being prepared.

LLeitech00:27Product release

AI score55

Zhuoyu Releases Pansing Commercial Vehicle Smart Driving Solution and End-to-End Large Model for Heavy Trucks

At its Zhuhai launch event on October 29, Zhuoyu unveiled Pansing, a smart assisted driving solution for commercial vehicles, and called it the world's first mass-produced end-to-end large model for commercial heavy trucks. The solution reuses the passenger-car foundation model and is trained on over 3 billion kilometers of passenger-car data and 400 million frames of heavy-truck data. It comes in two versions, PDS 4 and PDS 5, with the latter supporting highway and urban NOA. Zhuoyu says Pansing can reduce heavy-truck fuel consumption by 4.5%-7.5%, and XCMG, Shacman, Sinotruk, FAW Jiefang, Foton and Dongfeng Commercial Vehicle have all partnered with it.

Same story · 2 reports

#Product updates#English text is being prepared.#English text is being prepared.

LLeitech00:27Opinion

AI score58

The Foldable iPhone and Touchscreen MacBook Pro Have Arrived—How Much Pressure Are Developers Under to Adapt?

Leitech analyzes Apple's recently launched foldable phone iPhone Duo and first touchscreen MacBook Pro, arguing that the new interaction rules bring significant adaptation pressure to developers. iPhone Duo moves the navigation bar and toolbar to the side of the screen and, based on iOS 27, designs an equal-ratio split-screen mode for two apps, but after launch, many third-party apps still display only a small window with black bars on both sides when unfolded. The article cites developer feedback saying that adapting a native app may take only one to three hours, while an app developed with .NET MAUI took 19 hours and still had not completed all adjustments; another developer used AI Coding to complete adjustments to more than 50 interfaces in three days.

#Industry news#English text is being prepared.#English text is being prepared.

TTechCrunch AI00:23

AI score65

Musk Tightens Grok Bot Usage Limits Due to Excessive Costs

Musk confirmed on X that the tightening of Grok Bot's weekly usage limits was not a malfunction, but rather because the cost of the service was too high relative to revenue. Some users reported that weekend usage jumped directly from as low as 50% to 100%. SpaceX's filing last month stated that AI already accounted for 52% of the company's revenue, and its renamed Super Intelligence division achieved a profit of $431 million.

Same story · 4 reports

#Industry news#Agents#xAI

EElevenLabs Blog00:23First-party

AI score31

ElevenLabs releases SFX v3 API: supports multi-segment prompts to generate sound effects

ElevenLabs has released the SFX v3 sound effects model in the Sound Effects API, with the model ID eleven_text_to_sound_v3. Compared with SFX v2, it follows multi-instruction prompts more strictly and handles vocals and music better. The new model automatically determines sound effect duration based on the prompt, and duration_seconds can be used to specify the duration. prompt_influence defaults to 0.3, and increasing it can make the output more closely match the prompt. Loop generation currently only supports eleven_text_to_sound_v2.

#Product updates#Speech#English text is being prepared.

TTomasz Tunguz blog00:23Opinion

AI score55

Tomasz Tunguz proposes four economic questions for the four-layer AI stack

Tomasz Tunguz divides the AI stack into four layers: applications, harness and context, models, and data centers, and argues that each layer faces an unresolved economic question. He notes that in the same week, OpenAI and Anthropic models scored higher in Cursor than in their respective native harnesses; input context can account for up to 98% of all tokens; and a better harness can reduce the cost of the same answer by 71% and raise gross margin from 38% to 75%. At the data center layer, AI will require $4 trillion in new debt within five years, the Ornn B300 index has risen to $12.29 per GPU hour, and it has been growing at a compound monthly rate of about 20% since June.

#English text is being prepared.#Reasoning#English text is being prepared.

TTesting Catalog00:16Model releases

AI score74

Meituan releases the LongCat-2.5 multimodal model, with 1.6T total parameters and 49B active

Meituan has released LongCat-2.5, a natively multimodal agentic model with 1.6T total parameters and about 48B active, now available via API. The model targets long-horizon agentic tasks and strengthens visual understanding and agentic coding capabilities, covering complex coding projects as well as visual coding scenarios such as creative design and 3D. The company says it can be compared with GPT-5.5, GPT-5.6-Sol, Claude-Opus-4.8 and Claude-Opus-5.

Quoted postMeituan LongCat · @Meituan_LongCat

Introducing LongCat-2.5, our latest natively multimodal agentic model with 1.6T total parameters and ~48B active. Built for long-horizon agentic tasks, it brings stronger visual understanding and agentic coding to work ranging from complex coding projects to visual coding such as creative design and 3D tasks. Learn more about LongCat-2.5: https://t.co/ULLDweplqR

Same story · 11 reports

#Model releases#Multimodal#Agents

MMustafa Suleyman00:16

AI score34

Microsoft releases Humanist AI Code of Conduct and opens public comment

Mustafa Suleyman said Microsoft has released the Humanist AI Code of Conduct, outlining the company's thinking on AI development. He called this summer's cyber incidents an important warning, stressing the need to focus on why this technology is being built and how to ensure it is aligned and controlled in the interests of humanity and its values. The public comment period for the Code will end in 3 weeks, and Microsoft hopes to gather feedback.

#English text is being prepared.#Safety / alignment#Microsoft

TTibo00:15

AI score32

Tibo: dots will integrate GPT-6 Astra and GPT-6.1 Terra

Tibo said the team is expanding the family of frontier models supporting dots, including GPT-6 Astra and GPT-6.1 Terra, with more models to come. He said Terra brings speed improvements and a solid personality upgrade that will help dots reach more people.

#Product updates#OpenAI

RRohan Paul00:15

AI score66

Anthropic Updates IPO Filing: Third-Quarter Operating Loss of $567 Million

Anthropic updated its IPO filing, disclosing third-quarter results for the first time, with operating profit swinging from $974 million in the second quarter to an operating loss of $567 million. The main reasons are $5 billion in higher-cost short-term computing power and $2 billion in non-cash expenses for employee charitable matching gifts; computing power as a share of revenue rose to 58%, adjusted operating profit rose from $2.2 billion to $2.5 billion, but the adjusted profit margin fell from 19% to 15%. The filing shows total computing power commitments of $895 billion, of which $447 billion is expected by the end of 2030 and $70 billion is due within 12 months; Google's minimum commitment rose to $121 billion (through July 2033).

Same story · 21 reports

#Industry news#Anthropic#Google

RRohan Paul00:15

AI score30

Sam Altman: AI Intelligence Has Surpassed Top Scientists

Sam Altman said at the opening ceremony of ADIPEC in Abu Dhabi that at the beginning of this year, AI was mostly less capable than the smartest scientists, and now it is mostly smarter than them; previously, AI was mostly unable to take on white-collar work, and now it can complete many parts of it. He also emphasized that this statement is rough, that performance is unstable, and that it does not hold true in all respects.

Same story · 3 reports

#English text is being prepared.#OpenAI

XX.PIN00:14

AI score17

China Tech Weekly: CXMT Trial-Produces HBM3E, Brain-Computer Interface Restores Vision

ChangXin Memory Technologies (CXMT) plans to start limited mass production of HBM3E within two months, with a yield of only 20–30%, compared with about 70% for SK Hynix, while performance remains close to the previous generation, HBM3. Chinese doctors implanted a brain-computer interface in a 66-year-old woman blinded by retinitis pigmentosa, bypassing the damaged retina to directly stimulate the visual cortex; her vision recovered to about 0.125 (about 20/160), and she can walk independently, recognize characters, and identify her grandson. Henan accounts for about 80% of China's synthetic diamond output and is shifting high-thermal-conductivity diamond toward AI chip cooling and semiconductor packaging.

#Industry news#English text is being prepared.#Embodied AI

AArena00:14

AI score36

Agent Arena adds cost-efficiency metrics per token and per minute

Arena.ai says that at the same intelligence level, costs can differ by 10x, output tokens can be 6x higher, and time taken can double. Beyond task cost, Agent Arena adds metrics that measure agent performance per output token and per minute of waiting time; real-world usage data shows that cost rises fastest with intelligence level, the returns from waiting longer are uneven, and the number of output tokens can hardly predict the bill.

#Product updates#Evaluations / benchmarks#Agents

AAyi AI Notes23:45Opinion

AI score52

Terence Tao criticizes the way AI proves mathematics: automated, predatory mining

In a recent interview, Terence Tao criticized the way AI proves mathematics, describing the chaos surrounding the Navier-Stokes equations as a kind of automated, predatory exploitation and mining. He believes mathematicians want AI to serve as an on-call supercalculator, not as a competitor fighting over results; the value of humans solving difficult problems lies in refining conceptual systems and theoretical tools through long-term derivation, whereas AI, relying on computing power and black-box logic to directly take the answer, leaves behind theoretical ruins that are hard to understand and impossible to build on. He emphasized that a correct answer does not equal scientific progress.

Same story · 2 reports

#English text is being prepared.#Reasoning#English text is being prepared.

LLarsen Cundric23:45

AI score15

Larsen Cundric: Kids Can Make a Game in a Day with an Agent

Larsen Cundric recalls his childhood dream of making games, and marvels that today kids only need to describe their ideas to an AI agent and can be playing the game they described by that same afternoon.

#English text is being prepared.#Agents