A quick roundup of developer news since the last digest.

AI & tooling

  • Google releases Gemini 4 Hydrogen: Hydrogen is the fast, low-cost model in the Gemini 4 family. It is built for everyday coding and agent work, and costs a flat $1.50/$7.50 per million input/output tokens across its full 1M-token context. Google says it runs much faster than Gemini 3.8 Flash and scores far higher on Terminal-Bench. It is available in the Gemini API, AI Studio, the Gemini app and Search. Google
  • Antigravity's Orchestrator switches models during a task: A new option in Google Antigravity hands work back and forth between Gemini 4 Carbon (planning and recovery) and Hydrogen (edits and tests) within one session, rather than routing the whole task up front. Google reports success rates close to Carbon alone, with 24% less quota used and 19% faster runs. Google Antigravity
  • Edge0 open-sources a 145B model built to run on a laptop: Edge0-145B and Edge0-120B ship with an inference framework. The company claims about 29GB peak memory and more than 80 tokens/s on compatible laptops, with coding and agent results it says match some cloud frontier models. These are the vendor's own numbers, so expect independent tests to follow. edge0
  • Forensic report links Chinese frontier models to Blackwell GPUs: By analysing published weights, researchers found numerical patterns that match Nvidia Blackwell hardware in six recent models from DeepSeek, Moonshot, MiniMax and Ant Group. Older models from the same labs match Hopper. Exporting Blackwell to China is restricted, and the method could become a new tool for checking where a model was trained. Unemployment Labs

Releases

  • Rust 1.100 stabilizes the never type: After ten years and five failed attempts, ! can now be named anywhere as a type. The release also stabilizes the core of the allocator API for Vec and Box, and adds a DropGuard helper for rollback-style cleanup. Cargo gains unused-dependency reporting and a min-publish-age setting that delays picking up newly published crate versions, which helps defend against supply-chain attacks. Rust Blog

Security

  • How a 22-year-old PostgreSQL bug became RCE: Mehmet Ince's write-up of CVE-2026-6460 shows how changing a field's type inside a composite value leads to a reliable memory-corruption exploit in PostgreSQL core. It is one of the flaws fixed in this week's PostgreSQL releases, and a good reason to upgrade if you haven't yet. Mehmet Ince
  • Apple details Secure Exclaves for ambient sensing: Apple describes new silicon-level isolation that processes camera and microphone input for features like HomeView and "Hey Siri". Raw audio and video never reach the main OS; only simple signals such as "enrolled user recognized" do. The design also gives users a tamper-resistant indicator of when the sensors are fully active. Apple Security Research
  • United Airlines blocks third-party AI agents: The airline will restrict personal AI assistants from booking flights and managing reservations. It cites the risks of sharing credentials and FAA rules that require a human to complete some steps. Banks have issued similar warnings, so agent builders should expect more sites to close their doors to agents. TechCrunch

Infrastructure & reliability

  • Large Microsoft 365 outage: Outlook, Teams, SharePoint, OneDrive, Copilot and Defender XDR went down worldwide. Microsoft blamed "a large number of servers being unexpectedly unavailable" without saying why. It is the second major Microsoft 365 incident since September. TechCrunch
  • GitHub's October availability report: GitHub had eight incidents in October. It reports that 140 of 151 production services now run in Azure Central US, and that it has more than doubled Kubernetes compute capacity. The report also describes the write outages and migration monitoring gaps it is still working through. GitHub Blog
  • Why Cloudflare chose QuePaxa for consensus: Cloudflare explains why its Meerkat control-plane service uses QuePaxa, a leaderless consensus algorithm, rather than Raft or Paxos. A single slow or unreachable leader can stall leader-based protocols on the volatile wide-area Internet. Cloudflare Blog

Industry

  • Nvidia fallout continues after Huang's exit: Reuters reports another affair with an Nvidia employee. Senior Democrats have opened an inquiry into how the administration handled Nvidia chip sales to China. Hugging Face's CEO has also responded to questions about Nvidia's commitment to open-source AI. TechCrunch
  • AI-generated apps are swamping Apple's App Review: Developers of iOS music apps describe long delays and puzzling rejections as a flood of AI-built apps hits the review queue. An estimated 10 to 20 new music apps now launch every day. Synth Anatomy

This digest is generated automatically by an AI agent from public news feeds. Every
item links to its original source; check the source before relying on any detail.