China Launches Qiusuo 4.0, First National AI Full Supply Chain Benchmark

On Oct. 13, 2026, version 4.0 of “Qiusuo,” an evaluation benchmark covering the entire artificial intelligence supply chain from computing power and model algorithms to intelligent systems and products, industry applications, and safety management, was officially released. It is said to be the country’s first national-standard AI evaluation benchmark system. According to reporting by TMTPost’s official account, the system provides unified testing procedures and evaluation criteria for makers of chips, models, and other components.

It is distinguished by real-world workloads from sectors such as power, healthcare, and steel. It aims to address the industry challenge that evaluation results cannot be compared across environments. Multiple pilot evaluations for the supply chain have already been conducted. Product validation has been completed in finance, power, and oil and petrochemical scenarios. It is said to have accumulated many industry adoption cases and served as a model.

Overview and Positioning of Qiusuo 4.0

Qiusuo 4.0 is a framework for consistently measuring AI from infrastructure to applications. Lightweight code tools for computing power, large-scale models, agents, and others have been released. They are intended for industry to quickly reproduce evaluations and conduct software-hardware co-verification. It turns abstract benchmark scores into practical selection criteria and supports efficient adoption of technology.

An official from the Information Technology Center of the China Electronics Technology Standardization Institute explained the significance of the release.

Turning abstract benchmark scores into practical selection criteria to support efficient adoption of AI technology

The remark was cited in TMTPost’s reporting. The release of the tools makes reproduction and comparison by third parties easier. The key point is that they can be used as a basis for procurement and adoption decisions.

Four Areas Covered by the Evaluation System

The evaluation system is divided into four major areas: foundational hardware and software, models and intelligent systems, safety management, and industry adoption. It consists of 16 types of evaluation tools. It covers underlying computing power, software-hardware coordination, intelligent computing cloud, and intelligent software engineering. It comprehensively covers large-scale models, agents, embodied intelligence, intelligent terminals, AI safety, and industry scenarios.

It is designed to incorporate real-world operations into test workloads. It reflects on-site conditions in sectors such as power, healthcare, and steel. It aims to narrow the gap between laboratory scores and field performance. It enables side-by-side comparison among suppliers and increases transparency in selection.

Validation cases in finance, power, and oil and petrochemicals are leading the way. Methods for translating industry-specific requirements into evaluation items have been accumulated. Going forward, the focus will be on expanding target industries and updating the standards.

Aim of ByteDance Safety Unit Spin-Off

On Oct. 13, ByteDance’s Seed announced in an internal memo the establishment of a new unit. Named Seed Model-Alignment and Preparedness, it will be led by Yan Lin, reporting to Seed head Wu Yonghui. Staff from the former Responsible AI team and part of the dialogue team will move to the new unit.

The new unit will focus on model safety and alignment. Its core mission is to align model behavior with human intent and safety constraints. It will prioritize improving the safety capabilities of agent models. Its tasks include preventing advanced risks such as CBRN.

Its work covers model specifications, safety evaluations, alignment training, protection systems, and risk monitoring. It will also collaborate with external research and third-party evaluations. Seed says safety and alignment have become core capabilities for large-scale model applications. The move builds an organization anticipating the spread of autonomously acting models.

Current Status of DeepSeek Mega Fundraising

On Oct. 13, details of the latest fundraising by large-model company DeepSeek were reported, citing people involved in the deal. The fundraising amount is said to have increased from the previously planned 50 billion yuan to about 100 billion yuan. This is the second fundraising round, with the target amount doubling in less than half a month. New and existing shareholders are said to account for about half each of planned investments.

The doubling of the fundraising size is said not to affect the overall valuation. The current pre-money valuation remains about 500 billion yuan. New investors include Shixi Capital, CPE, Legend Capital, Geely, and others. The notable feature is that industrial investors are at the center.

First-round shareholders such as CATL, Tencent, and IDG Capital are also expected to continue participating. Among them, CATL is said to be planning an additional investment of about 10 billion yuan. The funds are seen as going to secure computing resources and accelerate research and development. It is a flow of funds strengthening the domestic-circulation AI value chain.

Embodied AI Startup Funding and Listing Delay

Embodied AI startup Zhiyuan Innovation has launched a new fundraising round. The amount is about 1 billion to 1.5 billion yuan, with this stage said to be near completion. The latest valuation is said to be about 50 billion yuan. The news was reported Oct. 13, citing investors.

The company announced in July 2026 that it had started preparations for a Hong Kong listing. At the time, it was said it could list as early as August, with a target valuation of HK$40 billion to HK$50 billion. Under relevant Hong Kong Stock Exchange requirements, if a listing applicant completes a new fundraising after filing its application but before listing, it cannot begin trading until 120 days after completion of the fundraising unless a special waiver is obtained. The aforementioned investors say the new fundraising makes completing an IPO within the year impossible. Company representatives declined to comment on the new fundraising, valuation, and listing progress.

Analysis of Latest Semiconductor and Overseas

Developments

At a Kirin briefing on Oct. 13, Xu Songyu of Huawei’s semiconductor business disclosed the architecture. The subject is the logic-folded Tao chip Kirin 9050 Pro. It includes coordinated operation of the new-generation Lingxi central processing unit, Maliang graphics processing unit, Da Vinci architecture neural network processing unit, Kirin AI image signal processing unit, Balong AI communications unit, function-flow task scheduler, audio-video processor, security processor, and Lingxi microkernel. He also revealed that the next-generation Kirin 2027 chip is ready, with all test results meeting expectations. Cumulative shipments of Kirin-powered devices have exceeded 100 million units since 2023. The Kirin 9050 series has achieved mass production.

At the Giga Texas research factory in Texas, the Terafab concept was presented. Tesla and SpaceX aim to manufacture and test about 10 chip designs per year to accelerate development. They can quickly test chips, produce new photomasks, and manufacture revised versions. The aim is a system that does not require intermediaries.

Overseas, Character.AI was found to have filed for bankruptcy restructuring. It filed for Chapter 11 protection in the United States on the 12th local time. It is said to have agreed to sell some assets to Walt Disney for $20 million. Court documents put estimated assets and liabilities at between $50 million and $100 million each.

In the automotive sector, an explanation of inspection results for the MAEXTRO V800 brake pedal was released. It was inspected by a nationally certified third-party testing agency on Oct. 11, 2026. It met the industry standard QC/T 788-2018 requirement of less than 5mm displacement under a 1,500N longitudinal load. Even under a tightened test up to a 3,000N longitudinal load, no cracks or breakage occurred.

Move to Clarify Cross-Border Data Rules for

Securities Firms

The Securities Association of China has clarified risk management requirements for cross-border data at securities firms. Impact assessments are mandatory for overseas transfers of personal information. The rules require unified procedures for cross-border information flows at securities firms. They directly affect operations such as model training and customer information management.

The securities industry handles large volumes of customer and transaction information. Reviews and record-keeping are required for overseas transfers. In AI applications, managing the provenance of training data becomes important. This clarification will affect industry compliance costs.

Editorial Opinion

On short-term impact, Qiusuo 4.0 is expected to be incorporated into procurement specifications within the next three to six months. Validation cases in power and finance will serve as references, and comparable procurement will spread. DeepSeek’s large fundraising is seen accelerating the securing of computing resources. The independence of the safety unit is assessed to push up demand for pre-deployment reviews.

On the long-term view, the domestic-circulation AI value chain is expected to strengthen over one to three years. The evaluation benchmark will promote co-verification of domestic chips and models, advancing software-hardware integrated optimization. Safety and alignment capabilities are seen becoming conditions for product selection. Updates to the released tools will determine the maturity of the ecosystem.

On questions from the editorial team, the remaining issues are the scope of disclosure of evaluation results and the frequency of updates. Whether the doubling of the fundraising amount leads to monetization needs to be verified. Whether external collaboration on safety evaluation proves effective requires close attention. How to ensure international interoperability of the standards is also an issue.

References

• “【钛晨报】“求索”4.0上线！国内首个人工智能全产业链评测基准发布；重庆市委原副书记、市政府原市长黄奇帆被查；DeepSeek据称将完成1000亿元融资”, by 钛媒体官方账号 — 钛媒体, 2026-10-13T23:20:00.000Z (ARR)
• Source URL: https://www.tmtpost.com/8167674.html