This page is available in English: Home, Bio, Work, Blog, Contact, Privacy, Photo rights and the ticker. Some images may still contain Dutch text.
Archive From the archive of The Latest A.I. News
This is an archive page with the A.I. news of Thursday 29 October 2026, as it stood that day. The summaries have not been updated since, and a lot may have changed. Today's news is on The Latest A.I. News.
Last updated:
Mistral opens Large 4, Google and Microsoft report record numbers driven by AI, and a report on distillation by Chinese labs raises tension around OpenAI.
The model targets long-running agentic coding and security work and, according to Google, tops Artificial Analysis's Intelligence Index with a score of 59. Access is staged through a small group of defenders and the Fairwind program, with a broad release only after more guardrail work. The introductory price is $2 per million input tokens and $10 for output, doubling afterwards.
Mistral has published the weights of Mistral Large 4, a multimodal mixture-of-experts model with about 1 trillion parameters and a 1-million-token context window. The license is a modified MIT variant: commercial use is allowed, but model-as-a-service providers need a separate agreement. The API costs $0.70 per million input tokens and $2.10 per million output tokens. Most benchmark scores are self-reported or come from partners; Artificial Analysis measures a lower Terminal-Bench score than Mistral reports.
The touch sensor is built into the OLED panel and macOS translates touches into clicks. Apple says the M5 Max can run models with hundreds of billions of parameters on-device, and Siri AI (beta in macOS 27.2) is not available in the EU. Prices start at $2,899.
Microsoft disclosed Azure revenue in dollars for the first time: $33.2 billion last quarter, growing 48%. Its quarterly filing shows more than $10 billion of revenue came from OpenAI arrangements, roughly 11% of the total. The company posted its tenth record quarter in a row.
Alphabet reports third-quarter revenue of $127.6 billion, with Google Cloud growing 103% to $30.7 billion according to the company; capital spending in 2027 is expected to top $300 billion. Growth in classic search ads slowed, while Google brings ads in AI Search results to more countries.
Instead of choosing a model yourself, Copilot routes each request to a single model or a combination (draft plus quality check, or critique), across three effort tiers. GitHub says the top tier lands close to Claude Code with Opus 5.5 on a coding index at roughly half the cost per task.
According to CNBC, the two chipmakers are putting roughly $4 billion into Meta, which is betting heavily on its own AI infrastructure. This is firmer than the earlier separate figures from Fox and Digitimes.
According to The Information, picked up by Reuters, DeepSeek (V4.1 Flash) and Moonshot (Kimi K3.1 Pro) relied heavily on answers from OpenAI's GPT-6 Astra to train their latest models, a technique known as distillation. Their scores on browser tasks rose to near Astra's level at a fraction of the cost per task. OpenAI had not responded yet; the method has not been independently confirmed.
According to Reuters sources, the US Department of Justice is examining whether Nvidia's acquisition of Hugging Face would hand it a key distribution point for AI. The review is part of a longer-running probe of Nvidia. Nvidia expects the deal to close in the first half of 2027 and says Hugging Face will stay neutral; an investigation is not evidence of wrongdoing.
The team is to spot abuse and route voting questions to trusted sources through November 3. ChatGPT had already restricted photorealistic political deepfakes ahead of the vote.
The figure is potential exposure, not confirmed compromises. It fits a wider picture of AI speeding up vulnerability discovery and exploitation: Australia's ASD also reports 20% more incidents.
Anthropic makes smart reports, in-app surveys and pooled group budgets generally available for Claude Enterprise. Admins can review up to 28 days of a team's work grouped by workstream, with costs attached; privacy safeguards such as restricted topics and user attribution off by default are built in. A member stops spending when either their own or the shared budget runs out.
Agents get verifiable identities, access limits and quickly revocable permissions, with extensions to HashiCorp Vault. It responds to banks and companies finding unmanaged agents acting on accounts.
OpenAI and ServiceNow are widening their partnership. A ServiceNow plugin for ChatGPT is available immediately, with more functions due by the end of the year and security and voice features later. Separately, ChatGPT can now search iCloud mail and calendar through new plugins, connected with app-specific passwords.
The models run on an ordinary CPU without a GPU and, by Mistral's own account, lead three forecasting benchmarks; independent confirmation is still missing. The license is a modified MIT.
The model transcribes, translates and identifies speakers. It is part of Nvidia's series of open models and data.
Wabi, which now calls itself a personal AI agent, raises $55 million in a round led by IVP, with Andreessen Horowitz, Forerunner and the Amazon Alexa Fund among the participants. That values the company at $1.25 billion.
Deepgram is buying Daily, the company behind the open-source Pipecat framework for voice agents; the price was not disclosed. Both sides say Pipecat will stay open source and vendor-neutral. Deepgram is also launching the Real-Time Context Standard, a freely available protocol for voice agents.
The wired glasses cost $1,099 at an introductory price through December and ship in mid-November; Gemini answers questions and explains what is on screen. A separate puck does the processing.
The open model has 744 million parameters; the text-only variant runs quantized in about 191 MB.
The essay argues that 'polysemanticity' has often been misread and that superposition has little support in real networks. The author calls it a working hypothesis; the figures come from small or preliminary samples.
Claude Sonnet 5.5 scores highest, ahead of Claude Fable 5.1 and Opus 5.5. Agents mainly fail on unstated consequences and unverified assumptions, such as a couple's marital status.
According to Business Insider these are early talks with investors; the company has confirmed nothing.
An update to OpenAI's API specification revealed an unannounced image model, GPT Image 2.6, with a new markup option for editing pictures. That hints at a quick launch, but OpenAI has not said anything about it yet.
According to Kotaku there are multiple takedowns of ports of games including Fortnite, Minecraft and Call of Duty; the piece argues small indie developers are hit too. The article does not say which AI tools were used.
Next up: November 3: US midterm elections; by November 4: Moonshot releases open weights for Kimi K3.1 Pro
Every day has its own page. They stay up for three months.
Everything here is a summary of reports from other media, not original research. The label (Certain, High, Medium, Low) is my estimate of how reliable the source is. The summaries are generated automatically with AI and may contain mistakes; read the article itself for the details (the star marks the leading source). Sources: leading A.I. and tech sites and the official blogs of the A.I. companies.
