Maxwell smart: new Chinese embodied model tops global AI ranking on physical tasks

scmp.com Oct 1, 2026

Maxwell smart: new Chinese embodied model tops global AI ranking on physical tasks

Curated by Shubham · TechCapsules · Oct 1, 2026

AI-summarised brief · reviewed before publication

The Chinese Academy of Sciences’ Institute of Artificial Intelligence for Industries unveiled Maxwell, an embodied AI model that achieved a record 91.9 score on the Meta‑World benchmark, the highest ever attained. Meta‑World, created by Stanford, UC Berkeley, and collaborators, tests robots on 50 everyday physical tasks such as grasping, carrying, opening doors, and using drawers. Maxwell’s performance surpasses competitors from Physical Intelligence, Google DeepMind, Alibaba, Meituan, Carnegie Mellon, MIT, Cambridge, and Toyota Research Institute, positioning China at the forefront of robotic physical intelligence.

💡 Why It Matters

· Maxwell’s benchmark victory signals a leap in practical robot autonomy, reshaping expectations for industrial automation and competitive research worldwide.

AI Capsules Robotics

Read full article on scmp.com

SEMIFIVE, Mobilint Partner on LPDDR6/UCIe Robotics AI Chip

SEMIFIVE has signed a turnkey development contract with South Korean AI semiconductor firm Mobilint to create a robotics AI chip under Korea’s Ministry of Trade, Industry and Resources “K-On-Device AI Semiconductor Technology Development” program. The chip will feature LPDDR6, PCIe Gen6, and UCIe‑S interconnects, enabling high‑performance, low‑power AI computation in outdoor environments such as agricultural robots. The collaboration uses SEMIFIVE’s Spec Hand‑off model, allowing Mobilint to set performance targets while SEMIFIVE manages design, packaging, testing, and mass production. The partnership aims to strengthen Korea’s domestic AI semiconductor ecosystem and secure an early lead in next‑generation robotics platforms.

💡 By integrating cutting‑edge memory, interface, and chiplet technologies, the joint chip promises robust, autonomous AI processing for field robots, reducing reliance on cloud connectivity and opening new markets for Korean manufacturers.

microsoft.com · Oct 1

Accelerating human capacity across healthcare with AI

Microsoft Ignite 2026 will showcase how AI expands human capacity in healthcare. The conference highlights real‑world deployments: Intermountain Health’s Dragon Copilot streamlines EHR documentation, reducing clinician admin time; Premera Blue Cross migrated its Facets platform to Azure, boosting batch processing by 38%; Pierre Fabre’s PLA.I.GROUND platform enables researchers to summarize documents and accelerate code development. These integrations illustrate AI’s role beyond efficiency, enhancing patient interaction, decision speed, and scientific discovery across providers, payors, and life‑science firms.

💡 By automating routine tasks, AI frees clinicians and researchers to focus on complex care and innovation, directly addressing burnout and accelerating breakthroughs in treatment and operations.

gamesradar.com · Oct 1

“AI upscaling is coming to PS5”: Sony’s base console is getting its…

Sony announced that its Quick Spectral Super Resolution (QSSR) AI upscaling will be added to the PS5, extending the image‑enhancement feature previously exclusive to the PS5 Pro’s PlayStation Spectral Super Resolution (PSSR). QSSR uses a streamlined neural‑network architecture and a hand‑tuned implementation to boost detail and stability without the heavy computational load that limited PSSR to the Pro model. Sony released patches for Marvel’s Wolverine and Ghost of Yotei, demonstrating clearer backgrounds, sharper character models and more stable motion. Engineers from Insomniac and Sucker Punch praised the technology’s ability to reveal fine details in complex environments. The rollout marks the first software‑driven visual upgrade for the base console six years after its launch, and no further titles have been confirmed yet.

💡 QSSR gives millions of existing PS5 owners a Pro‑level visual boost without buying new hardware, potentially extending the console’s relevance as the market shifts toward AI‑enhanced graphics.

socialsamosa.com · Oct 1

Decoding OpenAI’s strategy to counter competitors with Dots launch

On September 29, 2026, OpenAI introduced “Dots” at its annual Developer Day in San Francisco, marking a shift toward autonomous, always‑on AI agents. Dots, powered by GPT‑6 Astra, operates independently on OpenAI’s cloud, managing continuous tasks such as updating sales proposals, revising project documents, and building software demos across more than 4,000 applications. Integrated with Slack and Microsoft Teams, the agents can be summoned for ChatGPT or deployed directly into corporate workflows. The launch follows Meta’s Muse rollout and Instinct’s valuation surge, positioning OpenAI to capture enterprise use cases before competitors expand.

💡 By moving from reactive chatbots to proactive background agents, OpenAI targets the growing demand for seamless, real‑time workflow automation in businesses, potentially redefining how enterprises integrate AI into daily operations.

techradar.com · Oct 1

This Steam tool lets you fudge Nvidia DLSS 5 to easily work…

Borderless Gaming, a Steam‑available utility for borderless‑window play, has added a community‑created effect that enables Nvidia’s DLSS 5 neural rendering on PCs without modifying game files. The developer explains that the tool routes DLSS 5 calls through its own BGFX pipeline instead of Nvidia’s DLLs, allowing the feature to run on GPUs beyond the officially supported RTX 5000 series, such as the RTX 3090, which can reach roughly 30 fps with the effect enabled. Users must install Borderless Gaming and then download the DLSS5‑NR v0.2.1 effect from GitHub, following the provided instructions. While the implementation is a workaround rather than native support and performance varies by title, it offers a low‑effort way for enthusiasts to experiment with the latest AI‑upscaled rendering technology on modern PCs.

💡 By bypassing Nvidia’s licensing constraints, the hack democratizes access to cutting‑edge AI upscaling, letting owners of older RTX cards experience next‑gen visual fidelity.

nextplatform.com · Oct 1

Build the right AI factory for your needs: partner for success

HPE announced its AI Factory portfolio, a tiered suite of infrastructure solutions built with NVIDIA technology to support the full AI lifecycle—from data ingestion to model deployment and continuous monitoring. The offering includes three configurations: HPE Private Cloud AI, a turnkey on‑premises platform for up to 256 GPUs; HPE AI Factory at‑scale, a multi‑tenant system that can manage hundreds to tens of thousands of GPUs for large enterprises and service providers; and HPE Sovereign AI Factory, which adds air‑gapped operation, strict data residency, and compliance frameworks for organizations handling sensitive information. Each tier emphasizes defining workloads first, then matching compute, networking, storage, and governance to the intended business outcomes. HPE’s executive Thierry Pienaar highlighted that purpose‑built, end‑to‑end AI infrastructure is essential for turning data into durable competitive advantage.

💡 By modularizing AI infrastructure, HPE enables firms to scale sophisticated models without over‑investing in generic hardware, accelerating time‑to‑value for sector‑specific AI initiatives.