THE LATEST

Latest news

All articles, ordered by their latest update.

Filter and search
A penciled composition plan sits left of a flat mountain-and-lake poster on a desk, with a pencil below; the artwork reads Plan, then render.

Canva’s first in-house image model plans a design before drawing

Canva launched Craft on October 22 and says its new research lab built the model around design layouts, text, and editability. The release also gives a rare look at the lab’s size, while its strongest rankings and quality claims still come from Canva’s own evaluation.

A set of model cards sits inside a cloud project frame with a private-preview gate in front and a separate purchase path beyond it.

OpenAI models join Google Cloud’s menu, but access is still invite-only

OpenAI and Google Cloud say OpenAI frontier models and Codex workflows are coming to Model Garden. The current access form labels the offer a limited, invite-only preview for approved development and internal testing, with general availability planned for later.

A shop worktable holds brand swatches, a starless feedback note with a speech bubble, receipt, spreadsheet and separate app symbols around a planning sheet beneath One shop, many tools.

Google turns Gemini’s small-business features into a connected workbench

Google’s October 20 update adds a business landing page, saved brand styling, finance-oriented tools, and more app connections. It builds on an earlier Google Business Profile integration rather than introducing business context to Gemini for the first time.

Meeting, phone, and chat symbols flow into a shared context layer beside documents and customer records, then connect to a separate AI tool.

Zoom’s context layer connects conversation history to AI tools, feature by feature

Zoom introduced a Unified Context Layer at Zoomtopia, combining meeting, phone, chat, and contact-center conversations with documents and connected business records. Search and some APIs are marked available now, while several note-taking and assistant connections remain listed as future releases.

Six model cards are moved from a Copilot selector into an archive tray while replacement cards appear behind an administrator policy panel.

GitHub Copilot deprecated six model choices on October 19

GitHub confirmed the retirement of six model selections across most Copilot experiences. The final notice names replacements, but availability still depends on plan, client, and administrator policy, so a model change should be tested in the surface where a team actually works.

A generic Marketplace message panel shows a saved pickup note beside a connector permission control and an activity log.

A Marketplace case shows what “Always allow” can mean

The Associated Press reported that Meta’s Muse accepted a Marketplace offer and sent a seller’s pickup address under a standing permission. The case is a specific account of how a connector, a saved message template, and a user’s expectations met in one real transaction.

Three Beam checkpoint folders sit beside a license sheet and a row of server racks, showing that downloadable weights still need substantial infrastructure.

Beam’s Apache 2.0 release opens the weights, not the operating work

Reflection published Beam 1.0’s 501-billion-parameter weights in three precision formats on October 16. The Apache 2.0 license gives teams room to inspect and adapt the model, but a very large checkpoint still needs serious memory, serving infrastructure, evaluation, and guardrails.

A multimodal agent processes a document, chart, video frame, and tool call through one model, with a separate smaller edge model shown alongside.

NVIDIA’s Nemotron 3.5 Super VL targets visual, tool-using agents

NVIDIA released a 120B-parameter model that accepts text, images, and video for workflows involving documents and tools. Its headline speed claims describe specific vendor tests, while an independent benchmark measures a different slice of the model’s performance.

A large model is divided across several linked workstation towers, with a separate cloud API route and a license document beside them.

Muse Spark Open is downloadable, but its Mac route takes a cluster

Meta released the model weights on October 15 and showed them running across linked Mac Studios. That demonstration is one way to host an 816-billion-parameter model; it does not make local operation a one-computer download or remove the license conditions.

Conceptual map of Singapore with an office marker connected to public-sector model evaluation, incident response, and business adoption icons.

Anthropic opens a Singapore office and signs an AI safety MOU

Anthropic’s October 14 announcement pairs a local office with a government cooperation agreement on model evaluation, incident response, and responsible adoption. It is a framework for cooperation, not a new Claude plan or a published deployment contract.

Conceptual flow from an AI chat prompt to a task card connected to several generic app tiles, with a review mark beside the proposed result.

Gemini 4 Argon reaches the app, with task mode in tow

Google opened Argon to Google AI Pro and Ultra subscribers in the Gemini app and added a task mode connected to more than 40 apps. Access still depends on which Google product a person is using.

A workspace page connects to an external agent through a permission gate, with a separate skills library and version-history panel.

Notion 3.8 gives outside agents a route into workspace context

Notion’s October 13 release lets MCP-compatible agents read workspace context and update pages or task trackers. It also adds a Skills API and sharper database permissions, but the useful boundary is still the access granted to each connection and the person or agent using it.

THE WEEKLY EDIT

The week, made clear.

Stories worth understanding, brought together in one weekly edition.

Share this page

Instagram, TikTok & other apps

Copy the page link, then paste it where your chosen app supports links. You choose whether to send or publish.