Daily News Coverage: OpenAI Expands Astra Testing, Anthropic Launches Model Hardware Standard & Google Launches Gemini Omni 1.1 Flash
The CODEW Daily News Coverage | August 31, 2026
Good Morning! OpenAI's new Astra model is in internal testing with a major front-end capability leap, while the company has also terminated model supply to Cursor effective November 12 following SpaceX's acquisition. Google released Gemini Omni 1.1 Flash with 40-second video generation, Tencent open-sourced Hy4 Preview with 770 billion parameters, and Anthropic unveiled a new Model Hardware Standard (MHS) for connecting AI to physical devices. Meanwhile, the number of Gitea servers still exposed to a critical RCE flaw has surpassed 8,300, PaperCut issued another emergency patch for actively exploited zero-days, and ShinyHunters targeted McKesson in a patient data breach. Here's today's briefing from The CODEW's newsroom.
THE LEAD
OpenAI Expands Astra Testing, Cuts Off Cursor Model Supply as AI Competition Intensifies
What happened: OpenAI has expanded internal testing of its new Astra model, codenamed "mozaik-alpha-fdm," with developers reporting that the Max mode can generate 3D isometric maps and interactive web pages in a single zero-shot pass. Core breakthroughs include end-to-end multi-agent orchestration, persistent reasoning, and real-time self-correction. The model is expected to launch around September 3, directly competing with Anthropic's Fable 5.1. Simultaneously, OpenAI announced it will terminate direct model supply to Cursor effective November 12, following SpaceX's $60 billion acquisition of Cursor's parent company Anysphere. OpenAI cited an inability to confirm that SpaceX would adhere to service terms, and new models, including Astra, will no longer be provided. Developers will only be able to use their own API keys after the transition period.
Why it matters: The Astra testing expansion signals OpenAI's continued push to maintain its lead in the frontier model race, with the model's multi-agent orchestration and persistent reasoning capabilities representing a significant step forward in AI agent capabilities. The timing — a direct challenge to Anthropic's Fable 5.1 — underscores the intensifying competition between the two AI labs. The Cursor supply cut is a strategic move that highlights how the model supply chain is becoming politicized. Following Anthropic's similar move to limit Windsurf when it was being acquired by OpenAI, the neutrality of foundation model APIs is being eroded by upstream-downstream competitive relationships. This could accelerate the trend of AI companies building their own model capabilities rather than relying on external providers.
Who is affected: OpenAI, Anthropic, SpaceX/Cursor, developers relying on OpenAI's models, and the broader AI ecosystem.
What to watch next: The official Astra launch on September 3 and how it compares to Fable 5.1, as well as the response from the developer community to OpenAI's Cursor supply termination.
AI & Models
Google Launches Gemini Omni 1.1 Flash with 40-Second Video, Tencent Open-Sources Hy4 Preview
Google has released Gemini Omni 1.1 Flash, a new video generation model capable of reading up to 10 seconds of preceding video and extending it by 10 seconds per generation, with a cumulative length of up to 40 seconds. The model adds first/last frame control and supports up to 3 seconds of video reference input. A 360p draft mode delivers approximately 60% higher throughput than 720p at about one-third the cost, with final output supporting 4K resolution. The model is now available to developers via the Gemini API and Google AI Studio, with consumer access across AI Plus, Pro, and Ultra tiers globally. Separately, Tencent has released Hy4 Preview, an open-source model with 770 billion total parameters, 49 billion active parameters, and a 1 million token context window. The model covers software engineering, office analytics, game development, and scientific research. In blind tests by 163 internal experts across 203 engineering tasks, Hy4 Preview scored 2.99, slightly outperforming GLM-5.3 (2.92) and Kimi K3 (2.94), and can generate game prototypes through conversation. The model has also pushed the volume lower bound on the three-dimensional Blaschke-Lebesgue geometry problem to 0.41104.
Why it matters: Google's Gemini Omni 1.1 Flash represents a significant advancement in AI video generation, with the 40-second cumulative capability and 4K output pushing the boundaries of what's possible in AI-generated video content. The draft mode's efficiency improvements could make high-quality video generation more accessible to developers. Tencent's Hy4 Preview, with its 770 billion parameter count and strong performance against leading models, signals China's continued investment in large language models and its intention to compete globally in the open-source AI space. The model's application in scientific research — improving on a geometry problem — suggests that AI is increasingly being used as a tool for mathematical discovery.
Anthropic Launches Model Hardware Standard, Claude Code Desktop Gains Terminal Session Resumption
Anthropic has released a preview of its Model Hardware Standard (MHS), designed to connect AI to physical devices. The standard uses a unified translation layer to address instrument interface fragmentation, with safety thresholds written into the driver layer to intercept out-of-bounds commands. The reasoning layer is separated from the execution layer, offering both online command and offline burst modes to mitigate cloud latency. Carnegie Mellon has used MHS to accelerate experiments by approximately 3x, and QuEra reduced laser relock time from 150 seconds to 6 seconds with a 99.3% success rate. Danaher and AWS have joined the ecosystem. Separately, Anthropic has added a /resume command to Claude Code Desktop, allowing users to restore previous terminal sessions with full conversation history and context. The feature works bidirectionally with the existing /desktop command, and restored sessions still support /rewind checkpoint rollback.
Why it matters: Anthropic's MHS represents a significant step toward integrating AI with physical devices and robotics, moving beyond pure software applications. The separation of reasoning and execution layers with safety thresholds addresses critical concerns about AI controlling physical systems. The early adoption by Carnegie Mellon and QuEra suggests MHS could become a standard for AI-instrument integration. The Claude Code Desktop updates further enhance the developer experience, making AI-assisted coding more seamless across terminal and desktop environments.
Semiconductors
ChangXin Memory Begins Mass Production of LPDDR6, First Featured in Xiaomi 18 Fold
ChangXin Memory Technologies has announced mass production of its self-developed LPDDR6 memory, marking the first time a Chinese domestic memory manufacturer has achieved a global first in a high-end memory standard. The memory will debut in the Xiaomi 18 Fold folding flagship smartphone. The LPDDR6 features a peak data rate of 12,800 Mbps, a baseline rate of 10,667 Mbps, and a maximum capacity of 16 GB, matching the initial generation offerings from Samsung and SK Hynix. The company achieved this milestone just one year after the standard was finalized.
Why it matters: ChangXin's achievement represents a significant breakthrough for China's semiconductor industry, demonstrating that domestic memory manufacturers can compete with global leaders in cutting-edge memory technology. The rapid transition from standard finalization to mass production in just one year suggests that China's memory industry is maturing quickly. This could intensify competition in the memory market, potentially benefiting consumers through lower prices and increased supply, while also reducing China's reliance on imported memory chips.
NVIDIA RTX Spark AI PCs Sell Out as First Shipments from ASUS and MSI Booked Solid
NVIDIA's RTX Spark series PCs, developed in partnership with MediaTek, are scheduled for a fall launch. Microsoft and more game developers have recently announced updates and support for the platform. ASUS and MSI, among the first wave of manufacturers, have already sold out their initial shipments and are requesting additional allocations from NVIDIA.
Why it matters: The strong demand for NVIDIA's AI PCs signals robust consumer and enterprise interest in on-device AI capabilities. The sell-out of initial shipments suggests that the market for AI PCs may be larger than anticipated, which could accelerate the adoption of AI-powered personal computing.
Enterprise Software
Salesforce and Anthropic Launch Claudeforce with 37 Prebuilt Sales Skills for Pilot Customers
Salesforce and Anthropic announced Claudeforce on August 26, an expanded strategic partnership that brings Claude's reasoning capabilities together with Salesforce's data, workflows, business logic, actions, and governance. The first shipping component is a Salesforce plugin inside Claude carrying 37 prebuilt sales skills, available to select pilot customers ahead of an open beta scheduled for September 2026. The partnership establishes Claude as the default reasoning engine across the entire Salesforce ecosystem. Salesforce CEO Marc Benioff emphasized that AI is expanding platform value rather than displacing enterprise software. Salesforce also raised its full-year revenue and profit guidance after Q2 revenue climbed 11% to $11.35 billion, with shares rising approximately 13%.
Why it matters: Claudeforce represents a significant integration of AI into enterprise software, moving beyond simple chatbots to AI agents that can directly interact with CRM data and workflows. The partnership signals the end of model-agnostic enterprise AI, as Salesforce commits to Claude as its primary AI engine. This could accelerate AI adoption in sales organizations and put pressure on other CRM providers to develop similar AI integrations.
Cybersecurity
Over 8,300 Gitea Servers Still Exposed to Critical RCE Flaw; PaperCut Issues Second Emergency Patch
Shadowserver has disclosed that as of August 27, 8,393 public Gitea servers remain unpatched against CVE-2026-60004, a critical code injection vulnerability. The flaw allows attackers with repository write access to execute arbitrary shell commands via the diffpatch API, and because Gitea enables open registration by default, unauthenticated attackers can create accounts and repositories to trigger the exploit. Attackers have been observed deploying cryptocurrency miners on unpatched servers. Gitea released a fix in version 1.27.1 on July 27, and CISA has added the vulnerability to its Known Exploited Vulnerabilities catalog. Separately, PaperCut has issued a second emergency security patch for CVE-2026-81578 and CVE-2026-82078, which attackers are exploiting to bypass authentication and achieve remote code execution. Researchers found that the first patch could be bypassed in multiple ways, and actual exploitation has been observed in customer environments. The new patch covers PaperCut NG/MF versions 24, 25, and 26.
Why it matters: The large number of unpatched Gitea servers — over 8,300 — highlights the ongoing challenge of getting organizations to patch known vulnerabilities, even when exploits are active. The PaperCut situation is particularly concerning: the attackers found ways to bypass the initial patch, requiring a second emergency fix. This underscores the evolving nature of threats and the need for organizations to stay vigilant even after applying patches.
McKesson Patient Data Stolen by ShinyHunters; GiveWP Vulnerability Affects Over 100,000 Sites
Healthcare giant McKesson has been targeted by the ShinyHunters ransomware group, which claims to have stolen patient data. Meanwhile, Patchstack has disclosed a critical vulnerability (CVE-2026-82222) in the GiveWP WordPress donation plugin, affecting over 100,000 websites. The flaw allows attackers to create accounts via an unauthorized registration interface and trigger deserialization vulnerabilities through malicious serialized objects in the donation process. GiveWP released version 4.16.7.2 on August 27 to fix the issue.
Why it matters: The McKesson breach highlights the ongoing threat to healthcare data, which remains a prime target for ransomware groups. The GiveWP vulnerability, affecting over 100,000 websites, underscores the risks associated with third-party plugins in widely used platforms like WordPress.
Markets
Paradium.AI Begins Trading Under Ticker PAAI; LEAP 2026 Opens in Riyadh
The Arena Group, now operating as Paradium.AI, has completed its name change and will begin trading under the new stock ticker PAAI effective August 31. Meanwhile, LEAP 2026, one of the world's largest technology conferences, opens today in Riyadh and runs through September 3. Mannai Information Technology is among the exhibitors highlighting AI and digital innovation in sports, smart cities, and enterprises.
Why it matters: Paradium.AI's new ticker reflects the growing trend of companies rebranding to emphasize their AI focus. LEAP 2026 serves as a bellwether for technology trends in the Middle East and globally, with AI expected to be a central theme.
What to Watch
- OpenAI Astra launch (Sept 3): Whether the model delivers on its multi-agent orchestration promises and how it compares to Anthropic's Fable 5.1.
- Claudeforce open beta (Sept): How enterprises adopt AI-powered sales tools and whether this becomes a model for other AI-enterprise integrations.
- OpenAI-Cursor supply cut (Nov 12): How developers respond to the transition and whether other model providers step in to fill the gap.
- Gitea and PaperCut vulnerabilities: Whether the remaining unpatched Gitea servers are secured and whether PaperCut attackers find further bypasses.
- ChangXin LPDDR6 mass production: Whether Chinese memory manufacturers can sustain momentum and compete with Samsung and SK Hynix.
Reviewed by Erwin Castro
on
Monday, August 31, 2026
Rating: