The AI Wire ●

Perplexity released pplx-embed-v2-late on October 7, a pair of ColBERT-style embedding models for retrieving text, images and rendered document pages such as PDFs and slides from a single shared space. Both sizes are on Hugging Face under the MIT licence; a hosted API is planned but not yet live.

The line-up splits the work: a 0.6B model is built for fast, cheap live queries, including on edge hardware, while a 9B model is built for index quality. Because they share an embedding space, a corpus indexed with the 9B model can be searched with the 0.6B model, avoiding 9B inference cost on every query.

Instead of one vector per document, the models keep a compact 128-dimension vector per token and score matches with MaxSim. Perplexity reports 92.4% on the MADQA document-question benchmark for the 9B model, and the approach avoids an OCR step by encoding pages as images.

OpenAI has begun rolling out GPT-6 in ChatGPT alongside a new capability called Intelligent UI, which lets answers include interactive charts, forms, tappable buttons and diagrams instead of plain text alone. The rollout started October 7 for Plus, Pro, Business and Enterprise users, who get GPT-6 Sol, and reaches Free and Go tiers on October 8 with GPT-6 Luna. OpenAI says the feature is tuned for the more than 1.2 billion people who use ChatGPT each week.

Intelligent UI is powered by a library of native, streamable components and a compiler that processes the interface while the model generates it, so elements appear progressively rather than after the full response finishes. The company gave examples like an interactive recipe widget that rescales quantities by guest count, road-trip stops on a map, and on-demand mini tools: a savings calculator, a dinner bill splitter, even a small game played inside the conversation. GPT-6 also starts answering while it is still thinking, which OpenAI says makes responses arrive faster.

The Sol and Luna models debuted September 22 in ChatGPT Work, Codex and the API after the earlier GPT-6 Astra launch; this release moves them into the main Chat tab. OpenAI acknowledged the model's design judgment still needs work, closing its post with the line that 'instead of people adapting to software, software will adapt to people.'

OpenAI announced College Planner for ChatGPT for Teens, its default experience for users under 18. The tool pulls together application requirements, deadlines, tasks and financial-aid steps for the schools on a student's list into a single plan, alongside new flashcard and quiz tools and early usage data for the teen tier.

The initial US experience targets high-school students in grades 10-12 planning to attend a four-year college. OpenAI said hundreds of thousands of US teens already use ChatGPT each week to plan applications, track deadlines, research schools and understand financial aid.

The company said it plans to bring College Planner to more countries and to expand support to two-year colleges, technical colleges and trade schools over time. ChatGPT for Teens, which includes Study Mode, quizzes, learning visualizations and scheduled Study Hours with age-appropriate protections on by default, launched on August 18, 2026.

OpenAI's Thibault "Tibo" Sottiaux said on October 7 that Codex and ChatGPT Work together have reached 40 million active users, a milestone he shared as GPT-6 began rolling out in ChatGPT's main Chat tab. The announcement groups the coding agent and the workplace agent into a single adoption story as OpenAI broadens ChatGPT toward agent-led work. Sottiaux also said the company would add a banked usage reset to paid accounts.

The 40 million figure is company-reported and combines two products without a stated measurement window, so it signals reach rather than repeat use or either product's individual growth. For context, OpenAI said in June that Codex had passed 5 million weekly users, and in late August Sottiaux put ChatGPT Work at 20 million users.

The timing ties the usage claim to a wider product push: the October 7 GPT-6 release with Intelligent UI expanded the consumer surface on the same day Sottiaux promoted the combined agent audience.

OpenAI has rolled out a Meetings plugin for the ChatGPT desktop app, its answer to dedicated meeting assistants like Granola. The plugin listens in on virtual and in-person meetings through the microphone and system audio — no awkward bot participant joins the call — then combines the transcript with the user's notes and connected-app context into a personalized summary with suggested action items.

The generated meeting notes are saved as a page in ChatGPT Space, where they can be reviewed, edited and shared, and the suggested actions can be handed to ChatGPT Work to draft follow-up emails or update project plans. OpenAI displays a consent prompt requiring users to inform participants they are being recorded; audio is deleted from both the Mac and OpenAI's servers once notes are generated.

The catch is availability: the plugin is Mac-only for now, with iOS, Android and Windows support promised later, and it sits in beta for Pro and Business users. It is a clear land-grab for the meeting-notes market just as ChatGPT Work keeps adding workday features.

At Wednesday's Windows and Surface event, Microsoft detailed new AI capabilities for Windows centered on Copilot: local file context, so agents can work with files on the device, and OS-wide actions. A new Windows Search experience brings thousands of actions directly to the taskbar — switching to dark mode, turning on Do Not Disturb, dimming the screen, arranging windows, sending a text message — and an opt-in integration with the new Copilot will surface quick answers inline. The new actions began rolling out to Windows Insiders on October 7.

On the developer side, agents already supporting Microsoft's containment model include OpenAI's Codex, GitHub Copilot, and LM Studio, with Anthropic Claude Code, Perplexity, and Manus planning support. Notably, Microsoft said Meta's Muse personal AI agent is coming soon as a native Windows app with integration. Microsoft Entra will soon distinguish agent activity from user activity in Microsoft Agent 365, with Intune policy support for managing agent process containers on Windows 11.

The push makes safety containment a central theme: Nadella pitched this as the 'birth of the next generation of Windows agentic platforms,' while acknowledging that proving AI agents can be safely contained on personal computers is one of the key challenges.

Microsoft held its first major Windows event in over two years on Wednesday in San Francisco, with CEO Satya Nadella and Nvidia CEO Jensen Huang on stage. The headline device: the Surface Laptop Ultra, a 'builder tier' laptop powered by Nvidia's RTX Spark chip that Microsoft positions as the best machine for running local AI models. The company claims it outperforms the MacBook Pro with an M5 Pro chip in several AI workloads and can run AI models exceeding 120 billion parameters locally, thanks to up to 128GB of unified memory.

Pre-orders opened immediately at $2,599, with shipping starting October 16. Other RTX Spark laptops from Asus, Dell, HP, Lenovo, and MSI were shown alongside it. Microsoft also announced a $5,999 Surface RTX Spark Dev Box for developers running open-weight models, shipping to U.S. customers in November, and deskside DGX Station-class systems for running frontier-class models locally later in 2026.

The event was as much about Nadella's bet that some work currently done in Azure's data centers can shift to powerful Windows machines in businesses and homes. Deskside systems powered by Nvidia DGX Station, which Microsoft says can run frontier-class models at more than a trillion parameters locally, are due on Windows later this year.

The laptops are now actually on sale. NVIDIA's October 17 launch post puts RTX Spark laptops from ASUS, Dell, HP, Lenovo, Microsoft and MSI in stores starting at $2,399 — $200 under the Surface Laptop Ultra's starting price — still with up to 128GB of unified memory and up to 1 petaflop of AI performance, with more systems to follow.

Microsoft made MAI-Code-1.1-Flash available to download and run locally, with 3-bit quantization and a 256K context window. The small coding model, already in production in GitHub Copilot since August, claims 25% faster token streaming and 25% fewer tokens per task than the original MAI-Code-1.

The company also announced experimental on-device access coming to the GitHub Copilot app, Copilot CLI and Visual Studio Code by the end of October. Running it locally requires workstation-class hardware — Microsoft points to machines like the Surface Laptop Ultra with 128GB of memory.

The move extends Microsoft's in-house model strategy into on-device territory. MAI-Code-1.1-Flash costs a fraction of its predecessor — $0.20 per million input tokens and $1.20 per million output — undercutting the Chinese coding models Microsoft had cited as competitive pressure when it launched the model in August.

Meta updated its Muse personal AI agent app on October 7 to support iPad, nearly a month after the mobile app's initial release, The Verge reported. The new version is designed around the iPad's larger display and multitasking, and Apple's App Store listing now identifies Muse as compatible with both iPhone and iPad.

The update also adds a wave of new connectors for work tasks: Canva, Dropbox, Figma, QuickBooks, GitHub, Zoom, Asana, Klaviyo, Granola and Notion. The listing describes Muse as a personal agent that can manage email, calendars, reservations, budgets, shopping, health and fitness data, and linked social and business accounts, subject to user approval for each action.

The iPad release closes out the mobile footprint for Muse, which launched on iPhone, Android and the web on September 8 and in Canada on September 18. Windows remains the last major platform waiting — Meta and Microsoft have said a native Windows app with MXC integration is coming soon.

Google launched a public site on Tuesday that lets anyone verify whether a piece of media was generated using AI. The tool had previously been available only to select journalists, media professionals, and researchers for testing since Google I/O last year, and is now open to everyone. Google says people currently make 1 million verification requests a day.

The site supports JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF, and GIF for images; MP4, MOV, and WEBM for videos; and WAV, MP3, OGG, FLAC, AAC, and M4A for audio. Detection is based on SynthID, the watermarking system Google introduced in 2023. Media from Google's Nano Banana, Veo, and Lyria models, as well as generation-capable tools like Gemini, Flow, ProducerAI, and Vids, all carry SynthID marks. OpenAI, NVIDIA, and Kakao also support SynthID, while Apple is said to be adding support soon.

Google has baked SynthID verification into the Gemini app and Chrome. These tools are not infallible, though — they often fail to identify content created by their makers' own models, and Microsoft and Meta run their own separate watermarking standards. Still, a free public checker is a practical consumer tool in an election of misinformation.

Google DeepMind, Meta and AI drug-discovery company Isomorphic Labs will jointly invest $300 million in the Virtual Biology initiative led by the non-profit biomedical research organization Biohub, The Verge reported on October 7. The project marshals $1.8 billion in combined funding, data, and computing and measurement resources.

The central goal is a large-scale biological dataset that AI can train on, leading to virtual cell models that predict how cells respond to disease, drugs and other interventions — so researchers can run first-pass experiments digitally before the lab.

The US government is participating too: the Department of Energy will invest more than $500 million over five years in national-lab supercomputers and imaging technology, while NIH links and standardizes its biological datasets for AI training. Big-lab competition is heating up in biology, and this is the largest coordinated bet yet on AI-simulated cells.

Google has rolled out Gemini in Chrome to Android users across India and brought the agentic Auto Browse feature to the country on both desktop and Android. Gemini in Chrome itself is free: it can summarize pages, answer page-specific questions, connect with Google apps like Keep and Calendar, and generate images with Nano Banana.

Auto Browse, which lets Gemini navigate websites and carry out multi-step tasks like booking parking, filling long forms, or ordering groceries, is gated to Google AI Pro and AI Ultra subscribers. The rollout is phased: Android support currently works on devices set to US-English or UK-English, while desktop Chrome is offered in eight Indian languages.

Google says Auto Browse is secure-by-design, with models trained to detect prompt-injection attacks, and that it shows a plan of the websites it will visit before starting a task, asking for confirmation on sensitive actions like payments.

Security company Cequence found traffic matching Meta's Muse agent at more than half of the customers it studied, within two weeks of Muse's September 8 launch — and most of those businesses would not have known. Muse presents itself as an ordinary Chrome browser rather than identifying as an AI agent, so to standard security checks it looks like a person browsing.

Cequence spotted the traffic through behavioral analysis, including a browser update that went from 0% to more than 90% of Muse traffic within days — a sign Meta hosts and updates these cloud browsers centrally. At the median customer, Muse traffic grew nearly sixfold within about two weeks of first appearing.

The finding landed alongside the launch of Cequence's Agent Trust capability, which verifies agents that present an identity and catches the ones that don't. With agentic browsers multiplying, the company argues businesses need both: stop out-of-bounds actions while still letting customers use their services through agents like Muse.

Anthropic launched Claude Haiku 5.5 on Wednesday, the third model in its Claude 5.5 family in the past month, ahead of a planned IPO. The new small model targets lightweight, repetitive applications such as document summarization, data extraction, live customer support, and browser automation, and can also serve as a budget-friendly sub-agent alongside Sonnet 5.5 or Opus 5.5. Anthropic calls it the cheapest, fastest, and most capable small model it has released.

The headline number is price: for prompts under 100,000 tokens, which account for roughly 90% of historical Haiku workload, the rate drops to $0.10 per million input tokens and $0.50 per million output tokens — a 90% cut that exactly matches OpenAI's GPT-6 Luna. Above that threshold, pricing quintuples to $0.50/$2.50, still about half of Haiku 4.5. Anthropic says average running costs are around 75% below Haiku 4.5, factoring in an updated tokenizer.

Haiku 5.5 is also the first Haiku with an adjustable effort setting, letting developers dial compute up or down to trade cost against intelligence. It ships with a one-million-token context window and a first for the Haiku line: built-in safeguards for a narrow set of high-risk cybersecurity requests, without impacting most everyday tasks. The model is live now on the Claude Platform, AWS, Google Cloud, and Microsoft Azure under the model ID claude-haiku-5-5. With this launch, the 5.5 generation rollout that began with Opus 5.5 in September is complete.

SemiAnalysis limit-tested every major plan tier and found that Claude plans offer about five times the API value of ChatGPT plans on mid-tier models, according to an October 6, 2026 update relayed via the explainx.ai blog's OpenAI usage roundup. The comparison is about what subscribers actually get per dollar of subscription, not marketing token claims.

For anyone running serious workloads through a subscription instead of the API, the gap is the practical thing to know before renewing. A ChatGPT Plus or Pro renewal versus a Claude plan is no longer a close call on mid-tier usage value, per the testing.

The numbers come from SemiAnalysis's testing as reported second-hand, so treat the 5x figure as their measurement rather than a vendor claim. It lines up with the broader shift: plan value is quietly becoming the comparison that matters as the labs meter more aggressively.