Today's Headlines

  • Google unveils Gemini 4 Carbon, its strongest model, and hands it to cyber defenders first
  • Anthropic expands its GSA deal, offering Claude to all levels of U.S. government at 40% off
  • Preferred Networks releases PLaMo 3.1 Prime, with Deep Research that also runs on premises

The day belonged to the question of who gets powerful AI, and when. Google will put Gemini 4 Carbon, a model built for cybersecurity and long tasks, in the hands of cyber defenders before it reaches the general market.

Governments and companies also gained ways to run AI on their own terms. Anthropic extended its government discount to every product, and Preferred Networks shipped a new Japanese model that can run inside a customer's own data center.

Today's Top Three

Google unveils Gemini 4 Carbon, its strongest model, and hands it to cyber defenders first

Google DeepMind announced Gemini 4 Carbon on October 28, calling it the company's strongest model.

Carbon sits above Gemini 4 Argon, which launched two weeks earlier. It ranks first, with a score of 59, on version 5.0 of the Intelligence Index from Artificial Analysis, a firm that benchmarks AI models. Google says its own teams have seen Carbon run for days on hard engineering tasks such as tracing and patching subtle bugs across large codebases.

Its edge is cybersecurity. Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for confirming that a flaw can be exploited, and 71.0% on CWE-bench for fixing the root cause. Google's Chrome team reported that Carbon did more than ten times better than another lab's large frontier model at producing high-quality patches.

The rollout is staged. Google is taking part in the U.S. government's voluntary pre-release model testing and starts this week with a subset of cyber defenders. Access then widens to trusted partners through its Fairwind Program, and developers, businesses and consumers get the model once Google has finished tuning its guardrails.

Carbon launches at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off.

Anthropic expands its GSA deal, offering Claude to all levels of U.S. government at 40% off

Anthropic said on October 27 that a new agreement with the U.S. General Services Administration gives government buyers 40% off Claude.

The discount covers Claude Enterprise, Claude for Government and the Claude API, and for the first time it applies across every product. Direct purchases qualify today, and purchases through cloud providers follow in the coming months. State, local, tribal and territorial governments are covered as well as federal agencies.

Anthropic estimates the discount will save the government $1 billion. Agencies make no minimum spend commitment and pay for the usage they consume.

About 500 U.S. federal, state and local government organizations now use Claude. Weekly active users of Claude for Government are up about eightfold since February, and usage across federal civilian agencies has grown about 20-fold in a year.

Among its examples, Anthropic says Utah gave Claude Code to 34 developers across 10 state agencies, and 71% reported saving 20 or more hours by the end of the four-week pilot. Agencies can buy at the discounted price through GSA's Multiple Award Schedule starting today.

Preferred Networks releases PLaMo 3.1 Prime, with Deep Research that also runs on premises

Preferred Networks (PFN), a Tokyo AI company, released PLaMo 3.1 Prime on October 27, the latest flagship of the PLaMo models it builds from scratch in Japan.

Its new Deep Research feature takes a research topic, draws up a plan, gathers and analyzes information, and writes a sourced report on its own in minutes to tens of minutes. It also runs on premises: connected to an internal document search system, it can research a company's own files inside a closed network.

PFN says post-training made the model better at saying "I don't know" when it lacks the facts and at reading the context it is given. Across eight hallucination-related benchmarks, scores improved by about 15% on average over PLaMo 3.0 Prime. The base model comes from joint research with Japan's National Institute of Information and Communications Technology (NICT).

A technique called speculative decoding makes the model 1.7 times faster than 3.0 Prime. In non-reasoning mode, it cuts output tokens by 90%, which lowers usage costs.

PLaMo 3.1 Prime is available through PLaMo Chat, the PLaMo API and on-premises deployment. The API adds a free plan with usage limits, and the Standard plan costs 60 yen per million input tokens and 250 yen per million output tokens. Through November 30, PLaMo Chat users can run Deep Research twice a day for free.

Other Developments

Models

Products

Research

  • Anthropic reported early results from using Claude in drug development with six partner organizations and said it has started drug discovery of its own. It acquired Mithril Medicine, a one-person company developing treatments for muscular dystrophy, whose founder used Claude to coordinate contracts and experiments with 11 outside labs and took one treatment from idea to promising results in mice in under 18 months. Early progress on accelerating drug development with Claude (Anthropic)
  • Researchers from Harvard University, Meta and other institutions played five scam scenarios, voiced by six AI voice models and by real people, to 4,100 U.S. adults. On average across all scenarios, 16.5% said they would comply or were unsure, and scams that played on emotions or relationships had three to more than five times higher odds than routine support scams. AI voice scam calls could fool one in six (ITmedia NEWS, Japanese)

Business

Policy

Source: Selected by the editors from the AI news inbox (collected October 29, 2026: 60 items, 12 primary and 48 secondary).