Meta has unveiled Muse Spark Open, a new open-weight AI model designed to compete with Chinese offerings from Alibaba and others. The model targets coding and complex reasoning tasks and can run on Mac Studios equipped with M5 Ultra chips. Apple and Meta collaborated to optimize the model for Mac hardware, improving MLX machine learning framework support for multi-device communication.
Running Muse Spark Open across multiple Macs creates a desktop supercomputer setup. Four Mac Studios can generate over 36 tokens per second, while two Mac Studios with M5 Ultra chips can hold an entire model in memory. This approach highlights a growing trend of using consumer hardware for AI workloads.
In other model news, Elorian launched Glimpse, an AI model that reasons with images by zooming in, cropping, or running code during analysis. Trained by a small team on a few hundred GPUs, Glimpse is smaller than frontier systems but reportedly outperforms GPT-6 Astra on scientific chart reasoning and video captioning. Pricing sits at $0.60 per million tokens in and $3 per million tokens out.
Kambi, a sports betting company, processes over 70 percent of betting activity across its network using AI. The system draws on more than 15 years of historical trading and pricing data, with models developed starting in 2021 and fully embedded by 2024. It handled all 104 FIFA World Cup matches with 100 percent uptime, supporting complex products like Bet Builders and player props.
African organizations are increasingly building technology solutions in-house. According to the 2026 CIO100 Awards, 38.2 percent of projects were completed without external vendors. The study covered nearly 1,300 organizations across 10 countries. AI use rose from 54.8 percent last year to 84.8 percent planning investment in the next 12 to 24 months. Small projects under $10,000 returned 3.6 times their investment on average.
Washington state is developing rules for government AI use, with decision receipts proposed for public services. Behavioral scientist Gleb Tsipursky recommends that these receipts explain which AI system was used, what information it considered, who the human reviewer is, and how to seek reconsideration. The proposal targets consequential decisions affecting public benefits.
LogicMonitor's Elevate event focused on AI-driven observability, highlighting the shift from reactive IT support to proactive service delivery. GlobalFoundries reduced alert noise by 98 percent using Edwin AI. Five pillars were identified, including enterprise data foundations and trust. The event underscored the importance of governed automation in enterprise environments.
A Dell Technologies and Vanson Bourne study surveyed 1,500 UAE business and IT decision-makers about digital transformation. It found that 82 percent of organizations are becoming more selective with technology spending, and 86 percent prefer a single strategic partner. AI adoption is growing, but 68 percent of leaders worry about regulatory compliance slowing initiatives. Also, 68 percent find their infrastructure unprepared for AI workloads.
Books are being scanned and destroyed to train AI models, according to a France 2 investigation. The practice raises concerns about preserving cultural heritage, as millions of physical books are digitized and then destroyed, risking the loss of rare second-hand volumes forever.
Welsh Cabinet Minister Adam Price argued that small nations like Wales can compete in the AI race. Speaking at Harvard Kennedy School's Growth Lab, he said success depends on five elements: power, data, skilled people, public consent, and firms willing to redesign how they work. Wales's health data bank was cited as an example of how small nations can leverage connectedness.
Key Takeaways
- Meta launched Muse Spark Open, an open-weight AI model for coding and reasoning that runs on Mac Studios with M5 Ultra chips
- Apple and Meta collaborated to run Muse Spark Open across multiple Macs, achieving over 36 tokens per second on four Mac Studios
- Elorian's Glimpse model reasons with images and reportedly outperforms GPT-6 Astra on scientific chart reasoning, priced at $0.60 per million tokens in
- Kambi processes over 70 percent of betting activity using AI, handling all 104 FIFA World Cup matches with 100 percent uptime
- 38.2 percent of African IT projects in the 2026 CIO100 Awards were completed without external vendors, while 84.8 percent plan AI investment
- Washington state proposes AI decision receipts for public services to explain system use, data considered, and human review processes
- GlobalFoundries reduced alert noise by 98 percent using Edwin AI, as LogicMonitor promotes AI-driven observability and proactive IT delivery
- 68 percent of UAE leaders worry regulatory compliance is slowing AI adoption, and 68 percent find infrastructure unprepared for AI workloads
- France 2 reports millions of books are being digitized and destroyed to train AI models, raising cultural heritage preservation concerns
- Wales's Adam Price says small nations can compete in AI through five elements: power, data, skilled people, public consent, and firm redesign
Meta launches Muse Spark Open AI model to rival Chinese rivals
Meta released a new open-weight AI model called Muse Spark Open. The company says it rivals Chinese models from Alibaba and others. The model is designed for coding and complex reasoning tasks. It can run on Mac Studios with M5 Ultra chips. Users need two Mac Studios to create a desktop supercomputer.
Apple and Meta create Mac-based AI supercomputers
Apple improved its MLX machine learning framework for M5 family chips. The updates support low-latency communication across multiple Macs. Apple and Meta worked together to run Muse Spark Open on Macs. Four Mac Studios can generate over 36 tokens per second. Two Mac Studios with M5 Ultra chips can hold an entire model in memory.
LogicMonitor highlights AI-driven observability trends
LogicMonitor held its Elevate event focusing on AI-driven observability. The event covered autonomous IT and governed automation. GlobalFoundries reduced alert noise by 98 percent using Edwin AI. Five pillars were identified including enterprise data foundations and trust. The goal is to move from reactive support to proactive service delivery.
Washington state considers AI decision receipts for public services
Washington state is developing rules for government use of AI. Behavioral scientist Gleb Tsipursky recommends requiring decision receipts. These receipts would explain what AI system was used and what information it considered. They would also identify the human reviewer and how to seek reconsideration. The proposal applies to consequential decisions affecting public benefits.
Books scanned and destroyed to train artificial intelligence
Millions of physical books are being digitized and destroyed to train AI models. France 2 investigated this controversial practice. The process risks causing rare second-hand books to disappear forever. This raises concerns about preserving cultural heritage while advancing AI technology.
Elorian launches Glimpse, an AI model that reasons with images
Elorian introduced its first AI model called Glimpse, which can analyze images by zooming in, cropping, or running code during reasoning. The model was trained by a small team on a few hundred GPUs and is smaller than frontier systems. Glimpse costs $0.60 per million tokens in and $3 per million tokens out. Elorian claims it outperforms GPT-6 Astra on scientific chart reasoning and video captioning. The company plans to focus on industries with large video datasets and aims to build the world's best visual reasoning model next year.
African firms increase in-house tech and AI investment
African organizations are building more technology solutions internally, with 38.2 percent of projects in the 2026 CIO100 Awards completed without external vendors. The study covered nearly 1,300 organizations across 10 countries. It found that 84.8 percent plan to invest in AI over the next 12 to 24 months, and AI use rose from 54.8 percent last year. Small projects under $10,000 made up 27.2 percent of submissions and returned 3.6 times their investment. However, only 38.6 percent of organizations had tested an incident-response plan despite growing cybersecurity concerns.
AI is changing espionage with dead drops and old methods
The rise of AI and digital surveillance is pushing espionage toward older techniques like dead drops and invisible ink. A dead drop is a hidden location where spies leave messages or devices for others to collect. AI-powered tools are making these methods more sophisticated. The use of AI in espionage is growing among military, intelligence agencies, and private companies. While AI helps spies gather information without being detected, it also makes it harder for them to remain hidden.
Kambi uses 15 years of data to power AI sports betting
Kambi processes over 70 percent of betting activity across its network using AI technology. The company combines AI with more than 15 years of historical trading, pricing, and outcome data. Its AI models began development in 2021 and were fully embedded by 2024. The system handled all 104 FIFA World Cup matches with 100 percent uptime. Kambi's AI supports complex products like Bet Builders and player props while improving pricing, risk management, and margins for operators.
UAE businesses face AI readiness and security challenges
A Dell Technologies and Vanson Bourne study surveyed 1,500 UAE business and IT decision-makers about digital transformation. It found that 82 percent of organizations are becoming more selective with technology spending, and 86 percent prefer a single strategic partner. AI adoption is growing, but 68 percent of leaders worry about regulatory compliance and say it has slowed AI initiatives. Also, 68 percent find their infrastructure unprepared for AI workloads. Cybersecurity is now a baseline requirement, with 90 percent expecting security integrated from the start.
Welsh Minister Adam Price Says Small Nations Can Win in AI Race
Adam Price, the Welsh Cabinet Minister for Enterprise, Connectivity and Energy, spoke at the Harvard Kennedy School's Growth Lab. He argued that small nations like Wales can compete in the age of artificial intelligence. He said success depends on five elements working together: power, data, skilled people, public consent, and firms willing to redesign how they work. Price noted that Wales's small size forces agility and collaboration. He pointed to Wales's health data bank as an example of how small nations can leverage connectedness.
Sources
- Meta releases Muse Spark Open, a new ‘open-weight’ AI model
- Apple’s big AI week ends with Mac + Muse AI supercomputers
- AI-driven observability: 8 Elevate insights
- State's AI roadmap needs to show how AI does its work for public
- Focus - How books are being scanned and destroyed to train artificial intelligence
- Elorian’s first AI model wants to reason with images, not just about them
- African firms build more in-house technology as AI investment grows
- Invisible ink and ‘dead drops’: How AI is changing espionage
- Kambi: How 15 years of trading data turned into an AI edge
- UAE businesses weigh AI readiness amid economic and security pressures
- Welsh Politician Adam Price Says Small Size Gives Nations an Edge in AI Race at Growth Lab Talk
Comments
Please log in to post a comment.