Latest news, developments and reporting on Large Language Models.
Meta released Muse Spark Open, an 816B open-weight MoE with 42B active parameters, 1M context, multimodal input, and deployment on two Mac Studios.
BMW's new-generation iX3 will use a China-specific cockpit AI jointly developed with Alibaba, built on the Qwen model and integrated by Banma.
A 2021 internal Anthropic document, exposed through a copyright lawsuit, shows the company planned coding agents, tool use, long context and agent safety more than a year before ChatGPT appeared, according to Leifeng.com.
NVIDIA and IIT Madras' AI4Bharat plan to release Nemotron-pretraining-Indic-v1, about 2.8 trillion tokens across 22 Indian languages, with an open GPU-accelerated pipeline for generating the data.
Apple CEO Tim Cook said the China beta of Apple Intelligence will launch later this month without Siri AI, while Sspai also reported new AI, security, gaming and keyboard-PC products from SpaceXAI, Microsoft, HP and Step.
Apple has won approval to bring its first Apple Intelligence features to users in China in a beta later this month, Tim Cook said in Shanghai, with Alibaba's Qwen model set to power the experience. DeepSeek, TSMC, Anthropic and Ramp also reported major developments.
Meta has released open weights for Muse Spark Open, an 816-billion-parameter model built for coding and long-running agent tasks, with a license restricting hosted resale and military use.
NVIDIA released Nemotron 3.5 Super VL, a 120B-parameter open multimodal agent model that activates 12B parameters per token for enterprise agents, plus two edge models.
The October 15, 2026 episode of 9to5Mac Daily covered OpenAI's claim that a former employee took confidential device files to Apple, iPhone Duo repair costs, Apple's Switch to Safari campaign, and Apple Intelligence's reported China launch without Siri AI.
Hugging Face has released Granite Switch 4.1 in 8B and 30B versions, combining a general decision adapter, a safety-focused Guardian adapter, and a speech recognition model on one endpoint with minimal extra VRAM.
Meta launched Muse Spark Open, an 816B open-weight model optimized for Mac Studio with Apple's MLX.
FTC subpoenas OpenAI and Anthropic for documents on safety incidents and possible consumer protection violations.
Bank of America hired 1,000 contractors and spent more than $100 million fixing software flaws uncovered by Anthropic's Mythos, CEO Brian Moynihan said. Other major banks are also spending heavily as advanced AI security tools expose vulnerabilities faster than teams can patch them.
Nvidia on Oct. 15 introduced the open-weight Nemotron 3.5 Super VL, a 120-billion-parameter model for always-on agents, plus two smaller Bolt edge models. Nvidia says the family targets faster, more customizable enterprise and local AI deployments.
Common Sense Media's Youth AI Safety Institute says ChatGPT for Teens missed parental alerts during simulated mental health crises and failed to redirect teens to help in more than 25% of expert-designed cases, and it urges OpenAI to limit the chatbot to adults until the problems are fixed.
At Open Source Summit Europe 2026 in Prague, openKylin introduced KART, a unified AI runtime, and a system-level agent execution environment, while detailing AI features in openKylin 3.0 and RISC-V optimizations.
Tmall's Double 11 starts Oct. 20 with 1 million brands and 72 million products; AI and flagship stores are key themes.
MacPaw has launched Eney, a proactive AI assistant for Mac and iPhone that works across apps, files and the system. It combines local and cloud AI models, the Mnemos memory system and writing tools, after development began in 2019.
Bank of America brought in 1,000 outside contractors to patch software vulnerabilities exposed by advanced AI models including Anthropic's Mythos, adding hundreds of millions of dollars in third-quarter costs, CEO Brian Moynihan said.
Hikvision won the 2026 CIIF Industrial Automation Award for its AI-powered ultrasonic inspection technology. At the Shanghai fair, it showcased more than 100 industrial digitalization products and launched over a dozen new products, including 3D AXI inspection equipment and large-model-based monitoring and safety systems.
A survey of 6,300 European online banking users found 26% had switched banks over poor digital services. Banks are using AI and generative AI to deliver real-time personalized recommendations, communication and support.
AI infrastructure engineers in China are seeing sharp pay gains, with top talent earning up to 8 million yuan a year as model makers chase GPU cost savings, a Leiphone report says.
Cook says China's first Apple Intelligence features are approved and will enter beta in late October, excluding Siri AI.
Shanghai AI Laboratory visiting scholar Xu Disheng and colleagues at Greater Bay Area University used the Shusheng-S2 science model in a human-machine collaboration to complete a proof of the three-dimensional Ghys conjecture. A Peking University team led by Dong Bin and Liu Ruochuan formalized and verified the proof with Lean and the SHEAF framework, according to Leiphone. The paper is on arXiv and the formal code is on GitHub.