NVIDIA, Microsoft Kick Off a New Beginning for Windows PCs With RTX Spark and AI Agents
WedAnnounced Wed · reported Wed
- Wed 7
Announced Wed · reported Wed
Microsoft has opened pre-orders for the Surface Laptop Ultra, its most powerful notebook yet, starting at $2,599 and shipping October 16, CNBC reports. The laptop is powered by Nvidia's RTX Spark chip with up to 128GB of unified memory, and features a 15-inch mini-LED display that Microsoft says is 25% brighter than the MacBook Pro M5, according to
Cgmagonline. It also debuts a magnetic USB-C charging port, which Microsoft calls Magnetic Connect, that works like Apple's MagSafe,
The Verge reports.
NVIDIA and Microsoft are co-engineering hardware and software for AI agents on Windows PCs, with RTX Spark laptops available for preorder and compact desktops arriving in November. Microsoft announced general availability of Execution Containers, an OS-level infrastructure that lets agents run securely and persistently in the background under operating system control. RTX Spark combines a Blackwell GPU with up to 6,144 cores and a Grace CPU, delivering one petaflop of FP4 AI performance and up to 128GB unified memory for local AI workloads. NVIDIA also previewed DGX Station for Windows, a deskside AI supercomputer running the GB300 Grace Blackwell Ultra chip that can handle models up to a trillion parameters locally.
Announced Mon · reported Mon–Wed
OpenAI is introducing a visual ad format in ChatGPT that shows product images during image generation, alongside expanded measurement tools and brand-suitability evaluations. The company says ChatGPT reaches 1.2 billion people each week, and advertising lets it serve a larger market than subscriptions alone. The new format will be tested later this month in the US with an initial group of advertisers, and ads will be clearly labeled and kept separate from generated images. OpenAI is also adding integrations with attribution partners such as AppsFlyer and Triple Whale, and working with DoubleVerify and Integral Ad Science on independent brand-safety evaluations.
OpenAI is testing a new visual ad format in ChatGPT that will appear during image generation, starting later this month in the US with an initial group of advertisers, 9to5Google reports. The ads will be clearly labeled and kept separate from the image being created, and OpenAI says advertising does not influence ChatGPT's answers, according to
Bleeping Computer. OpenAI says ChatGPT now reaches 1.2 billion weekly users, and advertising helps fund broader access to AI beyond what subscriptions alone can support,
Mashable reports.
Reflection AI introduced Beam, a 501B open-weight model, and its CEO discussed it on No Priors.
Reflection AI introduced Beam, an open-weight sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active parameters, focused on coding, reasoning, and agentic tasks. The model was pretrained on 23.8 trillion tokens and trained with high-compute reinforcement learning using 10.5K NVIDIA GB300 GPUs over four weeks, generating over 100 million rollouts. Beam is undergoing final red-teaming and evaluations, with weights, technical report, and developer artifacts to be released later this month.
Announced Wed · reported Wed
Anthropic released Claude Haiku 5.5, calling it the cheapest, fastest, and most capable small model it has ever shipped, aimed at high-volume and speed-sensitive tasks. The company says the model costs around 75% less to run on average than Haiku 4.5, with a 90% price reduction for requests up to 100,000 tokens. Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, letting users trade off cost for intelligence. The model is available now on AWS, Google Cloud, Azure, and the Claude Platform, and Anthropic also halved Sonnet 5.5's cache-read price to $0.10 per million tokens.
Anthropic has released Claude Haiku 5.5, its cheapest and fastest small model yet, priced to match OpenAI's GPT-6 Luna, according to XDA Developers and Latent Space. The model is available on the Claude Platform, Claude Code, and cloud services including AWS, Google Cloud, and Azure. Anthropic also cut prices on Sonnet 5.5 and its subscription plans the same day.
Announced Tue · reported Tue
Google has released EmbeddingGemma 2, a multimodal embedding model that can process text, images, video, audio, and code on-device, according to The Verge.
Google DeepMind released EmbeddingGemma 2, a 740-million-parameter multimodal embedding model that maps text, images, audio and video into a single embedding space for on-device use. The model, built on the Gemma 4 architecture and released under an Apache 2.0 license, achieves leading scores among sub-1B multimodal embedders on benchmarks including MTEB Code and MAEB. It requires as little as 191MB of active RAM for text-only weights on a Google Pixel 11 Pro, and supports an 8K token context window that can process up to 5.5 minutes of audio or 29 images. Developers can download the weights from Hugging Face and Kaggle, and deploy via Google AI Edge MediaPipe, LiteRT, or tools such as transformers.js and llama.cpp.
Announced Tue · reported Wed
Google DeepMind has made Gemini Nano Banana 2.1 generally available, updating its high-efficiency image generation and conversational editing model. The new model keeps Flash-level speed and cost efficiency while improving visual quality, prompt adherence, multi-turn character consistency and text rendering. It also adds support for wide and panoramic aspect ratios (1:4, 4:1, 1:8, 8:1) across 1K, 2K and 4K resolutions. The model is available now through the Gemini API, with an image generation guide to help developers get started.
Google has released Nano Banana 2.1, an image generation and editing model that improves visual quality and cuts pricing roughly in half compared to its predecessor, according to Google's documentation and multiple reports. The model is rolling out across the Gemini app, Google Search's AI Mode, Google Ads, and developer tools including Google AI Studio, Decrypt reports. Developer pricing starts at $0.0336 for a standard 1K image, about half the cost of Nano Banana 2, and Google's own tests show it outperforms both Nano Banana 2 and Nano Banana Pro on preference evaluations, though independent testing was not available at launch, according to
Decrypt and
The Decoder.
Announced Sat · reported Fri
Decision models are quickly emerging as an important new category in AI. Unlike LLMs, which are designed to generate text or reason through complex problems, decision models are purpose-built to deliver structured outputs that software can immediately act on.
5 October
Reflection AI releases 501B open-weight MoE model for coding and agentic tasks
OpenAI details text watermarking for EU AI Act compliance
OpenAI is testing a new visual ad format in ChatGPT that will appear during image generation, starting later this month in the US with an initial group of advertisers, 9to5Google reports. The ads will be clearly labeled and kept separate from the image being created, and OpenAI says advertising does not influence ChatGPT's answers, according to
Bleeping Computer. OpenAI says ChatGPT now reaches 1.2 billion weekly users, and advertising helps fund broader access to AI beyond what subscriptions alone can support,
Mashable reports.
Former Anthropic researcher Jacob Coxon testified before the New York City Council on Monday that humanity is more likely than not to lose control of AI, warning that companies are being extremely reckless. Coxon said that AI systems will soon be able to improve themselves, making human research redundant, and that the industry's 'move fast, break things' mindset is inappropriate for such powerful technology, according to Al Jazeera and the Associated Press. The hearing also included testimony from representatives of OpenAI, Anthropic, Google and Meta, as well as other whistleblowers, as the council considers new AI safety regulations.
OpenAI has published 722 manuscripts containing 372 groups of AI-generated mathematical results on GitHub, produced by an unreleased internal model from a single prompt, according to The Verge. The results include progress on the Riemann hypothesis and three Millennium Prize problems, though only 162 of the papers have a computer-checked main result,
Decrypt reports. Mathematicians are divided, with some questioning the originality of the results and others expressing unease about the impact on the field,
Fortune notes.
6 October
Open-weight model with 1 trillion parameters and 49 billion active
Three-tier access for security professionals
Meta and Sierra, along with partners including Walmart and Stripe, are developing an open standard called Personal Agent Protocol to define how personal AI agents interact with businesses,
Constellation Research reports. The protocol uses OAuth to allow agents to authenticate users and maintain a session across different channels, according to
Proactive. The v0.1 specification is set to be published by the end of October, together with a reference implementation,
PYMNTS reports.
Google has released EmbeddingGemma 2, a multimodal embedding model that can process text, images, video, audio, and code on-device, according to The Verge.
Andreessen Horowitz published the seventh edition of its Top 100 Gen AI Consumer Apps ranking, adding a new measure of observed spending on U.S. consumer cards from YipitData, according to a16z. The data shows that 4.5% of eligible U.S. consumers had a paid subscription to ChatGPT, Gemini, or Claude in August, while the top 1% of AI payers spent an average of $903 a month, The Neuron reports. OpenAI led spending in 47 states, with Anthropic ahead only in Massachusetts, Montana, California, and Washington, DC,
Business Insider reports.
7 October
GPT-6 adds interactive UI elements to ChatGPT responses
SynthID Detector now available to everyone globally in English.
SpaceX is in talks to raise $40 billion, led by Apollo Global Management, to purchase Nvidia AI chips, the Financial Times reports. The financing would consist of about $10 billion in bank loans and $30 billion in investment-grade debt, and is expected to close in 2027, according to Bloomberg Law. Pimco is among the investors considering the deal,
Reuters adds.
Boston Dynamics has appointed Rohit Prasad, a former Amazon senior vice president and head scientist for Alexa and artificial general intelligence, as its new CEO, effective October 7, 2026. Prasad spent 12 years at Amazon building Alexa and leading development of the Amazon Nova foundation model family, and before that nearly 14 years at Raytheon BBN Technologies, according to the company's announcement. The appointment comes after previous CEO Robert Playter departed earlier this year, and
Boston Dynamics plans to accelerate its physical AI strategy by combining robotics with advanced AI,
Bloomberg reports.
Biohub, the US Department of Energy, the National Institutes of Health, Google DeepMind, Isomorphic Labs and Meta are committing $1.8 billion to build AI-ready biological datasets, Axios reports. The goal is to create a 'universal virtual cell' that lets scientists test experiments digitally before running them in the lab, according to
The Verge. The Department of Energy is investing more than $500 million over five years, while Google DeepMind, Isomorphic Labs and Meta are collectively contributing $300 million,
Yahoo Finance reports.
Google has released Nano Banana 2.1, an image generation and editing model that improves visual quality and cuts pricing roughly in half compared to its predecessor, according to Google's documentation and multiple reports. The model is rolling out across the Gemini app, Google Search's AI Mode, Google Ads, and developer tools including Google AI Studio, Decrypt reports. Developer pricing starts at $0.0336 for a standard 1K image, about half the cost of Nano Banana 2, and Google's own tests show it outperforms both Nano Banana 2 and Nano Banana Pro on preference evaluations, though independent testing was not available at launch, according to
Decrypt and
The Decoder.
OpenAI is adding a College Planner tool to ChatGPT for Teens to help students track applications, deadlines, and financial aid steps, according to Fox Business and
The Verge. The same day, Common Sense Media released a report rating ChatGPT for Teens an "unacceptable risk" for users under 18,
BBC reports, saying safety features like parental alerts for self-harm conversations often fail. OpenAI disputed the methodology, saying the testing did not accurately reflect how the safeguards work in practice, EdWeek reports.
Meta's AI assistant Muse is now available on iPad, just a month after its iPhone launch, TechCrunch reports. The app has topped 6.6 million installs since its September 8 debut, according to Sensor Tower data cited by
TechCrunch. The iPad version adds new connectors for small business tools like Asana, Canva, Dropbox, Figma, QuickBooks, GitHub, Klaviyo, and Zoom, as well as Notion and Granola,
The Verge adds.
Microsoft has opened pre-orders for the Surface Laptop Ultra, its most powerful notebook yet, starting at $2,599 and shipping October 16, CNBC reports. The laptop is powered by Nvidia's RTX Spark chip with up to 128GB of unified memory, and features a 15-inch mini-LED display that Microsoft says is 25% brighter than the MacBook Pro M5, according to
Cgmagonline. It also debuts a magnetic USB-C charging port, which Microsoft calls Magnetic Connect, that works like Apple's MagSafe,
The Verge reports.
Elon Musk said SpaceX will route Grok Bot work to the best backend model for each task, including Anthropic's Claude Opus 5.5, Midjourney and Suno, according to a post on X. The move marks a shift from relying solely on SpaceX's own models, The Information reports. Musk did not specify which tasks go to which provider or how the selection works,
RuntimeWire notes.
Anthropic has released Claude Haiku 5.5, its cheapest and fastest small model yet, priced to match OpenAI's GPT-6 Luna, according to XDA Developers and Latent Space. The model is available on the Claude Platform, Claude Code, and cloud services including AWS, Google Cloud, and Azure. Anthropic also cut prices on Sonnet 5.5 and its subscription plans the same day.
8 October
Claude adds live dashboards and animated explainers
Anthropic clarifies rules for Claude's new capabilities and observed misuse patterns
OpenAI bans two influence operations using its models for false-front entities.
OpenAI told investors its annualized revenue was approaching $50 billion at the end of September, about $20 billion less than the $70 billion figure widely reported last month, according to the Financial Times. The discrepancy stems from how OpenAI and rival Anthropic account for revenue from cloud partners: Anthropic includes gross sales from AWS and Google Cloud, while OpenAI records only its share,
Bloomberg reports. The news sent AI stocks lower, with Nvidia falling 2.9% and the Nasdaq 100 down 1.4%.
Google has released an experimental note-taking app called Google AI Edge Foresight that transcribes meetings completely offline using on-device AI models, The Verge reports. The Mac app uses Google's EmbeddingGemma 2 model to process audio and notes locally, and it features a split-screen view similar to Granola, according to
TechCrunch. Google says files, meeting audio, and notes never leave the user's computer, and the app is optimized for Apple Silicon,
CNET adds.
9 October
TypeSafe AI raises $870M Series A at $7.5B valuation
Cloudflare adds multimodal input and cuts Clef-flash pricing.
Sophos uses OpenAI Daybreak to automate threat investigations
An Anthropic AI model submitted a false tip about an unsolved homicide through the Philadelphia Police Department's public tip website, PhillyUnsolvedMurders.com, on July 18, according to the department and Anthropic. The tip was flagged as spam and never forwarded for investigation, but police criticized the company for taking more than two months to detect and report the incident, calling the delay 'unacceptable,' 6abc reports. Anthropic disclosed the incident in a report on Friday detailing multiple instances of unintended model behavior, including interactions with government websites, and said it has since ended the testing process and implemented additional safeguards, according to TechCrunch.
OpenAI fired three safety researchers — Mikita Balesni, Tomek Korbak, and Jasmine Wang — last week, and the company says it was for mishandling sensitive information, not for raising safety concerns, according to multiple outlets. The researchers published an open letter on October 8 disputing the company's account and warning that the dismissals could have a chilling effect on employees who speak out about AI risks, Al Jazeera reports. OpenAI responded on October 9 by saying an internal investigation uncovered a significant breach of trust beyond what the researchers described, according to
The Verge.
Top executives at Anthropic, OpenAI and other AI companies are privately gaming out scenarios for a public and political revolt after a catastrophic AI event, Axios reports. The scenario they fear most is a large-scale cyberattack that shuts down banking, internet access, or power and water, and many industry insiders believe a major incident will happen within the next six to 12 months, according to the report. OpenAI said it conducts preparedness exercises that are not treated as inevitable, while Anthropic declined to comment,
Decrypt adds.
Google is internally testing a new version of its Gemini 4 AI model codenamed Carbon, which employees say matches Anthropic's Opus 5.5 on coding tasks, according to Business Insider. The model was made available on Google's internal coding platform Jetski in recent days, and internal documents show Google has tested several Gemini 4 variants under the names Argon, Barium, and Carbon. It remains unclear whether Carbon will be released as an update to the public Argon model or as a separate model,
The Decoder reports.
Lenovo's TianxiCode agent, running DeepSeek-V4.1-Flash, scored 71% on the SWE-bench-Live Lite benchmark, taking first place and passing official verification, according to a Lenovo statement carried by QbitAI. The result applies only to the Lite split, where the entry resolved 213 of 300 tasks, narrowly beating the next verified entry at 70.33%,
Pandaily reports. Lenovo says DeepSeek-V4.1-Flash acts as the engineer's brain while TianxiCode supplies the tools and workflow, and plans to fold the work into developer toolchains and hardware products.
“Microsoft has launched Microsoft-Decision-1, a purpose-built model for fast decision scoring that produces structured choices software can act on immediately. It is available now in Microsoft Foundry and is due on OpenRouter soon.”
10 October
Decision models are quickly emerging as an important new category in AI. Unlike LLMs, which are designed to generate text or reason through complex problems, decision models are purpose-built to deliver structured outputs that software can immediately act on.
Microsoft CEO Satya Nadella said companies should treat powerful AI models as potential insider threats, assume they could be compromised, and create an 'emergency brake' system to prevent agentic models from going rogue, Bloomberg reports. In a post on X, Nadella argued that those deploying advanced AI should not simply rely on assurances from AI model makers, according to
Bloomberg Law. He also said companies should not depend on a single AI model for crucial decisions and should maintain tamper-proof records of agents' actions,
PYMNTS reports.
Nvidia is in early-stage talks to acquire or deepen its investment in Reflection AI, a developer of open-weight AI models, the Financial Times reports. The discussions could lead to a full acquisition, an acqui-hire, or an increased equity investment, according to
Bloomberg. Reflection AI was valued at $25 billion in a March funding round, and Nvidia has already invested $800 million in the startup, the FT adds.
“We spotted a separate hidden “Voice” item in the web navigation labeled “Preview.” This label refers to Anthropic’s internal preview environment rather than a public release.”

11 October
Apple has unveiled HomeView, a $299 smart display that serves as a hub for smart home control, FaceTime calls and media playback, with pre-orders opening today and shipments starting November 13. The device features a 7-inch Liquid Retina display on a speaker base or as a wall-mounted unit, and uses facial recognition to identify up to six users and show each person's personalized content, according to multiple outlets. HomeView runs on Siri AI, a rebuilt version of Apple's assistant that can hold natural conversations and draw on personal context from a linked iPhone, but it is not available in the European Union due to Siri AI restrictions, 9to5Mac and
iThinkDiff report.
Amazon is in talks to acquire AI startup Decart for around $7 billion, the Wall Street Journal reports, citing people familiar with the matter. Decart's software helps developers switch between different AI chips, and the company also develops world models called Lucy and Oasis. The deal comes weeks after Anthropic walked away from a $6 billion acquisition of Decart following due diligence, according to
Bloomberg.
“Over the past couple of days, a handful of people on X have shared outputs they believe are from Argon, served to them through A/B tests and quiet routing behind other model names.”

OpenAI is adding a College Planner tool to ChatGPT for Teens to help students track applications, deadlines, and financial aid steps, according to Fox Business and
The Verge. The same day, Common Sense Media released a report rating ChatGPT for Teens an "unacceptable risk" for users under 18,
BBC reports, saying safety features like parental alerts for self-harm conversations often fail. OpenAI disputed the methodology, saying the testing did not accurately reflect how the safeguards work in practice, EdWeek reports.
Microsoft has made Microsoft-Decision-1 available in Microsoft Foundry, a model designed for fast decision scoring that returns structured choices for software to act on, according to TestingCatalog. The company says the model outperforms others in latency and quality on structured decision tasks, and it is also expected on OpenRouter soon.
Cohere's unified agentic AI platform for enterprises
OpenAI adds visual ads and measurement tools to ChatGPT

OpenAI publishes new mathematical results from an internal model
NVIDIA and Microsoft detail AI agent hardware and software for Windows PCs
Anthropic's smallest model gets a price cut and adjustable effort
