AI news · 10 stories

AI news for Oct 30, 2026

Mistral releases Large 4 open weights for self-hosted AI · Gemini adds inline file previews and editing · Prime Intellect releases faster kernels for MoE training

News card: Mistral releases Large 4 open weights for self-hosted AI
Hugging Face

Mistral releases Large 4 open weights for self-hosted AI

Mistral Large 4 weights became available on October 29, with text-and-image input and deployment support across the Hugging Face ecosystem. Its open-weight license includes restrictions for model-as-a-service offerings, making the terms important for commercial deployments.

Read at Hugging Face
News card: Gemini adds inline file previews and editing
Google

Gemini adds inline file previews and editing

Google announced in-chat creation, previewing and editing of documents, spreadsheets, presentations and code on October 29. PDF preview and editing also work on mobile, while the other formats are supported on the web, reducing the need to move between apps.

Read at Google
News card: Prime Intellect releases faster kernels for MoE training
Prime Intellect

Prime Intellect releases faster kernels for MoE training

Prime Intellect introduced Prime Mega MoE on October 29, adding a backward pass to DeepSeek’s fused kernel so it can support training. The company reports up to 1.34 times faster end-to-end training against its Prime-RL baseline by overlapping GPU communication and computation.

Read at Prime Intellect
News card: XREAL opens AURA orders with Android XR and Gemini
Google

XREAL opens AURA orders with Android XR and Gemini

Google and XREAL launched AURA wired Android XR glasses on October 28, opening orders in the US, Canada and South Korea, with Japan following November 2. Gemini can use context from the real world and on-screen content, bringing its assistance into spatial computing.

Read at Google
News card: Amazon reports AWS growth alongside heavy AI investment
Amazon

Amazon reports AWS growth alongside heavy AI investment

Amazon’s October 29 results showed AWS quarterly sales rising 39% to $46 billion. Trailing twelve-month free cash flow fell to a $42.4 billion outflow, primarily reflecting AI infrastructure investment, highlighting the cash demands behind cloud expansion.

Read at Amazon
News card: GitHub makes Agentic Workflows generally available
GitHub

GitHub makes Agentic Workflows generally available

GitHub made Agentic Workflows generally available on October 28, turning natural-language Markdown instructions into Actions workflows powered by coding agents. Read-only defaults, sandboxing and validated outputs provide controls for automating tasks such as triage, documentation and CI analysis.

Read at GitHub
News card: GitHub brings Agent Merge to cloud pull requests
GitHub

GitHub brings Agent Merge to cloud pull requests

GitHub integrated Agent Merge into its pull-request pages on October 28, allowing Copilot to fix failing Actions workflows and merge conflicts in a cloud environment. Paid Copilot users can combine it with auto-merge while retaining repository merge requirements.

Read at GitHub
News card: GitHub previews shared MCP connectors across Copilot
GitHub

GitHub previews shared MCP connectors across Copilot

GitHub announced a public preview of verified MCP connectors on October 28, launching with 26 services. A single connection works across Copilot clients and cloud sessions, reducing repeated setup for accessing external work context.

Read at GitHub
News card: GitHub open-sources Replica for AI-assisted prototyping
GitHub

GitHub open-sources Replica for AI-assisted prototyping

GitHub released Replica under the MIT license on October 28, letting teams use natural-language requests to create interactive prototypes grounded in their existing design systems. An optional self-hosted hub supports sharing and feedback, helping teams test product ideas earlier.

Read at GitHub
News card: GitHub previews code scanning for AI application risks
GitHub

GitHub previews code scanning for AI application risks

GitHub introduced public-preview scanning for AI applications on October 28, covering risks such as indirect prompt injection, sensitive-data disclosure and excessive agency. For enabled repositories with GitHub Advanced Security, pull-request scans flag risky agent code paths and provide mitigation guidance.

Read at GitHub

Get the next episode first

Shorts and deep dives on AI system design land on @AI.JoinDev first. Subscribe, or follow the RSS feed.