The importance of inclusion is highlighted when discussing technological advancements, as no progress should occur without accessibility. Data provided by the U.S. National Institute of Standards and Technology (NIST) reveals that the error rate in facial recognition systems can be 1% for individuals in their 20s, rising to 5% for people aged 70 or older.
Based on the last Census, conducted in 2022, Brazil had 30,936,186 citizens between 20 and 29 years old and 14,292,869 people aged 70 or older. In a simulation where every Brazilian would undergo facial recognition once, it is estimated that a 1% error rate among young people aged 20 to 29 would result in about 309 thousand failures. The 5% rate for those over 70 would project approximately 715 thousand failures.
The mandatory use of facial recognition for granting, maintaining, and renewing federal social security benefits, as well as for unlocking INSS benefits aimed at contracting payroll loans, may constitute an accessibility barrier for the elderly population.
Analyzing the digital literacy difficulties faced by the elderly, it is clear that facial recognition goes beyond simply looking at the screen. The process requires following several instructions, such as finding a location with a uniform background, aligning the front camera to the face, adjusting the device distance, and executing various confirmations before sending the image. These steps are considered exhausting and not intuitive, even being questioned by younger people accustomed to daily technology use.
This problem is not new. Previous debates on algorithmic racism have already proven how insufficiently representative training bases can harm the performance of these technologies in certain groups. This issue resurfaces when noting the scarce representation of aged faces in databases, which decreases system accuracy. Similar difficulties are felt even more acutely by Black, Indigenous, and trans people, according to the Brazilian Consumer Defense Institute.
To ensure accessibility, it is essential that technology companies and organizations implementing these solutions invest in both updating the systems and improving the user experience. This implies providing more didactic guidance, clear instructions, and a truly intuitive step-by-step guide, enabling individuals with less digital familiarity to complete the procedure without needing external assistance.
In a country characterized by its vast racial, ethnic, generational, and gender diversity, the discussion about technology access must necessarily take into account the diverse profiles that make up the Brazilian population. Technological advancement can only be considered a real advance when it is capable of encompassing color, race, ethnicity, gender, and age, something that, in current practice, is still far off.
ZTE has once again won the WAIC SAIL Star award thanks to its OEX supernode, which features a zero-cable design, the ability to connect 128 Graphics Processing Units (GPUs) in a single rack, and a comprehensive artificial intelligence infrastructure covering cloud supercomputers and Nubia agent phones.
The OEX supernode received the WAIC SAIL Star award for the second consecutive year. It solves the complex engineering challenge of connecting hundreds of GPUs at the scale of data centers while minimizing cable complexity, signal loss, and latency. Thanks to its orthogonal design, where computing racks and switching racks are connected vertically, the OEX supernode provides full interconnectivity for 128 GPUs in a standard rack without internal cables or a rear panel. This significantly reduces signal paths, lowers connection costs, and simplifies maintenance compared to traditional designs.
The architecture allows scaling up to 16,000 GPUs. ZTE developed proprietary chips for in-rack switching that simultaneously support the CLink protocol, overseen by China's Ministry of Industry and Information Technology (MIIT), and Broadcom's SUE protocol. This ensures GPU compatibility independent of the specific chip supplier, preventing vendor lock-in. The five-chip co-probing technology, which includes a Central Processing Unit (CPU), internal server interconnect, inter-rack communication, network card, and DPU, allows the system to dynamically select optimal configurations for different types of GPUs and workloads, making the OEX supernode a flexible platform adaptable to the rapidly evolving domestic GPU ecosystem.
ZTE's exhibition at WAIC was divided into four zones: AI Infrastructure, AI Applications, AI Smart Home, and AI Agent Smartphone. These zones presented a unified concept of AI—from computational power generation to end-user consumption. The infrastructure zone showcased the OEX supernode alongside the AIDC computing power cooperation architecture and the computing network structure.
The applications zone demonstrated the NewStart AIOS operating system, designed for industrial, home, and personal scenarios, as well as the Co-Claw agent corporate platform. This platform is used internally by ZTE to build an AI-managed organization and is replicated for the industrial, railway, mining, metallurgical, and energy sectors.
In the smart home zone, ZTE's products for home terminals were displayed, maintaining global market leadership for five consecutive years with over 100 million annual shipments, including 3 million AI-enabled devices. The agent smartphone zone featured the Nubia NaviX Ultra—the world's first production-ready agent smartphone, developed in collaboration with Doubao from ByteDance, which debuted at WAIC. Ni Fei, President of Nubia at ZTE, emphasized that the agent phone category represents the lower half of AI, performing tasks rather than just answering questions, which requires thorough engineering verification of all user pain points, native OS integration, stable manufacturing experience, and privacy assurance.
In addition to the exhibition, ZTE announced achievements in medical AI: generating medical reports with a tenfold increase in efficiency, creating HTA reports in 1–5 minutes, and assisting in cardiovascular disease diagnosis with 96% accuracy based on 10,000 clinical cases. Three humanoid robots were also presented: Xingzai, weighing 30 kg for commercial services; the heavy-duty Xinqing for industrial manufacturing; and the wheeled Xinwang. All operate on the Star Core automotive chip with the dexterous Xingqiao hands, achieving precision below 0.2 mm.
Alibaba has introduced the Qwen3.8-Max model, which boasts 2.4 trillion parameters and is released under an open license. Alibaba claims this model ranks second in strength after Anthropic Claude Fable 5. This release comes amid a highly active month for Chinese open AI, during which four major labs launched trillion-parameter models.
The Qwen3.8-Max model was available for preview on the Alibaba TokenPlan and Qoder platforms, with the full weights set to be released publicly. This announcement marks the culmination of an outstanding month for Chinese open AI, as several models such as Moonshot AI K3 with 2.8T, DeepSeek V4 Pro with 1.6T, and MiniMax M3 Pro, reportedly reaching 2.5–3T, were presented over the past few weeks.
The significance of this event goes beyond just the number of parameters. Unlike previous Chinese open releases that focused on price-to-performance ratios, Qwen3.8 enters a completely different competitive environment. The performance gap between cutting-edge open-source models and closed models has significantly narrowed. Six months ago, trillion-parameter models were exclusively proprietary; now, they have become the standard for open releases.
Three distinct strategies have emerged: DeepSeek focuses on aggressive price reduction, offering market disruption; Moonshot AI emphasizes transparency by publishing full benchmarks, including weaknesses; and Alibaba employs an ecosystem lock-in strategy by integrating Qwen3.8 directly into its developer toolkit.
Integration into the toolkit is a hallmark of Alibaba's strategy. Qwen3.8 is not distributed as an isolated API or model weight but is embedded into TokenPlan for AI-powered coding sandboxes, into Qoder as a native IDE, and into QoderWork for low-code enterprise collaboration. This approach, where the model resides within tools, creates an ecosystem lock-in effect that pure weight releases cannot provide, positioning Alibaba as a platform layer rather than just a model provider.
The broader industry context is also important. Daily token consumption in China reached 140 trillion in March 2026, representing a 1000-fold increase over two years. The availability of trillion-parameter open models directly influences this consumption economy. Even if independent benchmarks for Qwen3.8 do not fully confirm the claim of being second only to Fable 5, the existence of a 2.4T parameter model without licensing costs fundamentally challenges the pricing power of closed, advanced models. When open capabilities reach an 80-point capability scale, the premium for the remaining 20 points becomes increasingly difficult to justify, especially for price-sensitive developer and enterprise segments.
The release of Qwen3.8 concludes a defining moment for Chinese open AI. The concentrated launch of K3, DeepSeek V4, MiniMax M3, and Qwen3.8 within one month collectively raised the baseline capability level of open source to levels unimaginable a year ago. For the global AI landscape, the conclusion is clear: Chinese open models are no longer just catching up; they are setting the pace in accessible advanced AI, redefining the competitive dynamics between open and closed ecosystems and challenging long-held assumptions about the necessity of proprietary model economics in the AI industry.
Moore Threads demonstrated its concept of three AI factories at the WAIC 2026 event: a model training factory, a token generation factory, and an agent production factory. These factories constitute a comprehensive production infrastructure necessary for the era of tokenomics, where the demand for computing power grows exponentially due to the consumption of each other by AI systems.
All three factories are built on the Kuae computing cluster and the company's patented MUSA architecture. This architecture uniquely integrates AI acceleration, 3D graphics rendering, physical simulation, and video codecs into a single chip.
The most significant achievement is the EvoPhys-World model—the world's first 5D model fully trained on Moore Threads' MTT S5000 GPUs. This model ranked first in the Stanford WorldScore global ranking for world generation for 37 consecutive days. A research group from Peking University utilized the Kuae cluster with full support of the MUSA software stack, demonstrating that domestic GPU infrastructure is capable of supporting the training of advanced global models with quality, stability, and scalability comparable to international standards. This confirms the potential of domestic GPUs for complex AI tasks previously considered the prerogative of NVIDIA.
The technical foundation is the MTT C256 supernode, which uses a single-layer Scale-up network architecture. It provides full interconnectivity for 128 GPUs in a standard single-node width chassis, with the possibility of expanding up to 256 GPUs in a combined configuration. The orthogonal structure eliminates issues of bandwidth loss and high latency found in traditional multi-level networks. The model's FLOPs utilization rate reaches 60% for dense models and 40% for MoE models, effective training time exceeds 90%, and linear scaling efficiency reaches 95%, while training accuracy matches major international computing cards. The Kuae cluster successfully demonstrated stable training on 1000 cards for the multimodal model Wan2.1 and completed end-to-end training of the base MoE-236B model from scratch.
Moore Threads also introduced two AI-oriented terminals: the MTT AICUBE home AI center, positioned as a family smart hub, and the MTT AIBOOK laptop for mobile AI tasks. Combined with the MusaCoder code model, the MUSACODE AI programming agent, and the Moore Academy, the company presented a complete stack of solutions—from cloud supercomputing infrastructure to edge deployment and end-user terminals. This allows the company to position itself as the only Chinese GPU manufacturer that simultaneously covers both corporate and consumer markets with a complete scenario coverage strategy.
Beyond the three factories, a broader conclusion is that the Chinese ecosystem of domestic GPUs has reached a level of readiness for industrial application. BAAI used the Kuae cluster with MTT S5000 to complete the full training process of the generalized embodied brain RoboBrain 2.5. The BEVFormer benchmark on Moore Threads infrastructure achieved throughput on one and two machines comparable to leading international GPUs. Since daily token consumption in China exceeded 140 trillion in March 2026, increasing 1000-fold in two years, the ability to train, run, and deploy AI workloads on domestic GPU infrastructure has transformed from a theoretical possibility into an operational reality.
