Last checked October 14, 2026 · News roundup
Today in brief: Apple finally put a price on its Siri AI home hub. HomeView costs $299, preorders opened on October 13 and it ships November 13, but not in the European Union. The same day, Google moved Gemini 4 Argon into the Gemini app for paying subscribers, and Anthropic explained why some Claude API requests are now blocked before Claude answers, and billed anyway.
Below are nine stories, AI first, then SaaS and business. Each one covers the change, the people it hits, the price and your next step. If you are tracking renewals, our month-by-month SaaS price increase tracker lists the October changes in one place.
Part 1 · AI
Apple HomeView price and release date: $299, shipping November 13
Apple announced HomeView on October 13 through press releases rather than a keynote, MacRumors reports. It is a 7-inch square smart display built to show off Siri AI, priced at $299 in the US. Preorders opened the same day and deliveries start November 13, according to 9to5Mac. Colors are silver, starlight, rose pink and space gray.
What you get, per Tom's Guide: HomeView recognizes up to six household members by face and gives each a personal home screen. Siri AI can pull a recipe out of your messages or email, answer longer questions, handle calendar tasks and run smart home devices. There is a 12MP Center Stage camera for FaceTime and a built-in Safari browser.
Two catches. Personal answers come from Apple Intelligence-enabled iPhones at home on the same Wi-Fi, so an older iPhone limits what the hub can do. And the wall-mounted version starts at the same $299 but needs a mount sold separately. HomeView also skips the EU because Siri AI isn't available on iOS there; Apple says it is "working constructively with the EU" and gave no date.
What to do: confirm your iPhone supports Apple Intelligence before you preorder, and add the mount to your budget if you want it on a wall.
Gemini 4 Argon in the Gemini app: which Google AI plans get it
This is an update to our Gemini 4 Argon price and access explainer. On October 13, Google said Google AI Pro and Ultra subscribers can now use Gemini 4 Argon in the Gemini app. Android Authority adds that it also reaches AI Mode in Search for those plans. Developers get it on the paid tier of the Gemini API, in AI Studio, Android Studio and Google Antigravity.
The API price has not moved: $2 per million input tokens and $10 per million output tokens at the introductory rate, with cached input 95% cheaper. Google's own plans page lists AI Pro at $19.99 a month and AI Ultra from $99.99 a month, with a $199.99 tier that carries higher usage limits. Google's post does not mention the free tier or the $4.99 AI Plus plan, so don't count on Argon there yet.
Mac users get one extra: Google brought Antigravity's agentic coding interface into the Gemini macOS app. According to The AI Insider, researchers can also apply for six months of the top Antigravity tier at no charge, subject to eligibility and capacity.
Claude API blocked requests: Anthropic now bills them and bans distillers
Anthropic published a report on October 12 about "distillation," where rivals feed Claude prompts at scale and train their own models on the answers. Most of it is about other labs, but three changes touch paying API customers.
- Blocked requests cost money. Since September 24, requests that Anthropic's safeguards stop before Claude responds are billed, in areas that include distillation.
- A new classifier screens traffic on Claude Sonnet 5.5, Opus 5 and 5.5, and Fable 5 and 5.1. Anthropic puts its false-positive rate below 0.1%, which still means some legitimate requests will be stopped.
- Repeat offenders get throttled or banned. Accounts that keep trying to extract reasoning face stricter usage limits; accounts tied to distillation are banned. Anthropic says tens of thousands have been, often within two hours of a campaign starting.
Anthropic also changed how reasoning is exposed: for API accounts created on or after August 31, 2026, preserved thinking can't be edited out of prior context, and the company says some developers had to update their integrations. It names Zhipu and Moonshot among the labs involved.
What to do: if your app makes heavy, repetitive calls that ask Claude to spell out its reasoning, check your billing for blocked-request charges, log every refusal, and raise a support ticket if a normal workload gets flagged.
Linus Torvalds used Claude Code on the Linux kernel: what the patches do
Linus Torvalds pushed a branch called noref-filemap-read-with-llm to his kernel tree, Hardware Busters reported on October 12 after Phoronix spotted it. The eight patches change how the page cache handles small reads, and the cover letter says Claude Code wrote the code, tests and first draft under his direction.
The gains are large on one workload. On a single 4-core laptop with eight threads reading one file, kernel cycles per read fell from about 2,950 to about 725, and throughput rose from roughly 10 million to 30–33 million reads per second. It is not a free win: a 16-byte single-thread read into an unmapped page got slower, from about 4,150 to 6,550 cycles, and Torvalds calls the deferred-freeing piece "the part of this series that is least solved." Nothing has been merged.
Anthropic added its own engineering note on October 13. It replaced Git internally with a system called Scion after daily merges in its monorepo grew from about 2,500 in May to more than 30,000 in September. Scion is not released publicly, and Anthropic says it is no general-purpose Git replacement. If you'd rather have AI build a whole app than patch code, compare the options in our AI app builders hub.
Facebook "Your algorithm": how to choose the topics in your Feed and Reels
Meta said on October 13 that it is rolling out two AI tools that let Facebook users steer Feed and Reels. The first, called Your algorithm, is a space reached from the three-dot menu where you add and manage the topics you want to see, described in your own words. The second adds follow-up questions after you tap "Interested" or "Not interested," so Facebook learns what you actually liked or disliked.
Meta says large language models read the intent behind what you type and blend it with its usual ranking signals, and that recommendations can shift within the same session. Instagram already offers a similar topic control for Reels. Meta gave no countries or rollout dates, so it may not appear on your account yet.
Who it affects: everyday users get a cleaner way to fix a feed full of things they don't want. Brands and creators should expect organic reach to lean more on clear topic signals, so state what a post is about in plain words in the caption rather than relying on hashtags alone.
Part 2 · SaaS & business
Pluto Security raises $42M to watch the AI agents your staff build
Pluto Security has raised $42 million, TechCrunch reported on October 13: an $8.5 million seed round led by TLV Partners plus a $34 million Series A led by Jump Capital. The company monitors the AI tools employees use and everything bolted onto them, including skills, extensions, MCP servers, apps and agents. It flags risky components and can step in mid-session, either stopping an action or making the agent ask for confirmation, without banning AI outright.
Traction is early but real. Pluto has sold since April, has passed $2 million in annual recurring revenue and counts more than 30 enterprise customers, including Oscar Health, Workato, Sixt and AlphaSense. Its most useful finding for IT teams: about 60% of the risks it detects sit in vibe-coded tools built by non-developers, with finance the biggest source, then marketing and legal; developers rank fourth.
What it costs: Pluto hasn't published pricing. Its free Plutonium tool checks public AI skills, connectors and add-ons, and includes an MCP scanner. Even without buying anything, list which connectors your teams have plugged into Claude, ChatGPT or Copilot this quarter.
Anthropic and NTT Data: a $6.3 billion data center planned for Osaka
NTT Data is in the final stages of talks to build Anthropic's first dedicated data center in Japan, Nikkei Asia reported, as summarized by Data Center Dynamics on October 13. The site would be in Osaka Prefecture with up to 500 MW of capacity, at a cost of about $6.3 billion. The companies aim to finalize the deal by the end of fiscal 2026, with partial operations expected in 2028. Neither company commented.
For context, DCD notes that Anthropic has signed for at least 14.8 GW of compute across data centers and cloud providers since October 2025, and that it confidentially filed for a US IPO in June 2026. NTT Data already runs more than 150 data centers in over 20 countries and regions.
Who it affects: nobody's Claude plan changes today. For US businesses, the takeaway is scale: Anthropic keeps adding capacity abroad as well as at home, which matters if you depend on Claude for production work. Treat this as reported, not confirmed, until the companies announce it.
Reuters: Anthropic and OpenAI wanted a say over outside safety reports
Anthropic and OpenAI pushed for significant control over what outside evaluators embedded in their labs could publish, Reuters reported on October 13, citing two people familiar with the talks (summarized by The Next Web). Both labs worried about trade secrets reaching rivals. Ideas discussed included lab staff monitoring review teams and barring reviewers from working at two labs at once.
The timing is awkward. Last month, Dario Amodei said reviewers should be able to publish their main findings unedited, and OpenAI said much the same days later. Yet until at least October 5, both wanted tight control; by October 8, Anthropic's talks with some reviewers appeared to have moved on. Anthropic says reviewers it has worked with before set their own methods and publication decisions, while it keeps sign-off rights with new firms. OpenAI says it doesn't approve whole reports, usually just removes secret details, and lets reviewers disclose large cuts.
What to do: when a vendor cites an "independent evaluation" in a security review, ask who the evaluator was and whether they could publish without the vendor's approval.
Microsoft Maia 300 chip: what it means for Copilot capacity
Microsoft detailed Maia 300, its next in-house AI inference chip, in an Azure blog post on October 12. Each accelerator carries 288 GB of HBM4 memory with 18 TB/s of bandwidth, and a rack packs 96 of them into 167 kW. In that full configuration, Microsoft claims up to 7 times the rack-scale token throughput of Maia 200 and more than 2.5 times the performance per watt. The compute chiplets use TSMC's N2P process.
Microsoft says the chip is meant to add capacity for products such as Microsoft Copilot, GitHub Copilot and Microsoft Foundry. The post gives no regions, availability date or pricing, and promises more at Microsoft Ignite 2026.
Who it affects: nothing changes on your Microsoft 365 or Azure bill because of this post. Cheaper in-house inference is how cloud vendors protect margins on AI features, so watch whether Copilot limits loosen or prices hold when Microsoft shares more at Ignite.
FAQ: HomeView, Gemini 4 Argon and Claude billing
Is Apple HomeView available in Europe?
No. Apple says HomeView won't be sold in the European Union because Siri AI isn't currently available on iOS there. MacRumors reports it launches in eight countries on November 13. Apple says it is working with the EU on a path forward but has given no timeline for a European release.
Can free Gemini users try Gemini 4 Argon?
Not according to Google's October 13 post, which names only Google AI Pro ($19.99 a month) and Ultra (from $99.99 a month) subscribers for the Gemini app. Developers need the paid Gemini API tier, at $2 per million input tokens and $10 per million output tokens during the introductory period.
Does Anthropic charge for blocked Claude API requests?
Yes. Since September 24, 2026, requests that Anthropic's safeguards block before Claude responds are billed, in areas including distillation. Anthropic says its new classifier wrongly blocks fewer than 0.1% of benign requests. If a legitimate workload keeps getting blocked, log the requests and contact Anthropic support.
Did Linus Torvalds merge AI-written code into Linux?
Not yet. Torvalds published an eight-patch branch whose cover letter says Claude Code wrote the code, tests and first draft under his direction. It tripled small-read throughput on one laptop but slowed one edge case, and Torvalds calls part of it unsolved. It sits on a separate branch and has not been merged.
Sources (accessed October 14, 2026): MacRumors, Apple recap (Oct 13) · MacRumors, HomeView and the EU (Oct 13) · 9to5Mac (Oct 13) · Tom's Guide (Oct 13) · Google, Gemini 4 Argon in the Gemini app (Oct 13) · Android Authority (Oct 13) · The AI Insider (Oct 13) · Google AI plans · Anthropic, distillation defenses (Oct 12) · Hardware Busters (Oct 12) · Anthropic, Scion (Oct 13) · Meta Newsroom (Oct 13) · TechCrunch, Pluto Security (Oct 13) · Data Center Dynamics, citing Nikkei Asia (Oct 13) · The Next Web, citing Reuters (Oct 13) · Microsoft Azure blog, Maia 300 (Oct 12)
