โ† Blog Frontier Models AI Economics Policy & Education

Gemini 4 Finally Lands at #1, Anthropic's IPO Paperwork Lights a Political Fire, and Computer Science Enrollment Falls Off a Cliff

By Best AI Tool Team October 29, 2026 5 min read Last updated: October 29, 2026
Gemini 4 Carbon Launch Anthropic IPO Scrutiny and Computer Science Enrollment Trends
Share this article:

Disclosure: Anthropic, which makes Claude, appears in two sections below.

For ten weeks this blog has tracked one question: where is Gemini 4? Yesterday, Google finally answered. And the same 24 hours delivered the clearest picture yet of how much money the industry is betting on that answer, and what students and lawmakers think about it.

Gemini 4 Arrives, for Real This Time

On October 28, Google unveiled Gemini 4 Carbon, priced at $2 per million input tokens and $10 per million output tokens. Our October 2 post wrongly said the model had already shipped, and we corrected it on October 13. This launch comes five weeks after DeepMind's chief said post-training had begun and the model would arrive “much earlier” than year-end.

According to AI Weekly's summary of Google's announcement, Carbon ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59. It reportedly scores 88.2% on the SEC-bench Pro vulnerability-detection test, 71.0% on CWE-bench patching, and 54.8% on ExploitGym. It ships first to cyber defenders through the Fairwind Program, the same gated route Google used for its security-focused Flash model in September. Google also claims its Chrome team measured a tenfold improvement in patch quality and a 0.0% prompt-injection success rate in internal tests. These are launch-day claims; independent testing will take weeks. Explore how Gemini compares to frontier alternatives in our AI Comparisons hub.

Earnings Week: How Big Is the Bet?

  • Alphabet: Third-quarter revenue of $127.6 billion (up 25%), with Google Cloud up 103% to $30.7 billion. Its model APIs now process nearly 32 billion tokens a minute, four times last year's rate, and the Gemini app has reached 1 billion users. Capex this quarter was $51.4 billion; it raised 2026 guidance to $195โ€“205 billion and signaled 2027 will top $300 billion. Core search growth slowed to 12% from 19% in Q1, and Google is extending ads in AI Mode to ten more countries.
  • Microsoft: Quarterly revenue of $92.3 billion, Azure up 48%, and capex of $39.4 billion, below the $50 billion analysts expected, with about $175 billion guided for calendar 2026. Its filing also shows OpenAI-related arrangements generated $10 billion in revenue for the quarter, nearly 11% of the total, with $7.2 billion in receivables outstanding.

Anthropic's Paperwork Meets Congress

Representatives Greg Casar and Maxine Waters introduced a nonbinding “sense of the House” resolution urging that no exceptional federal support go to AI companies or their lenders that fail in a downturn. It leans on Anthropic's public IPO filing, which the resolution says shows $16.3 billion in first-half revenue, $80.7 billion in first-half losses, and $518โ€“807 billion in planned obligations over a decade. It also names Google, Amazon, Microsoft, Nvidia, Broadcom, and CoreWeave as exposed counterparties.

Caveats matter here. The numbers come second-hand through a political document reported by The American Prospect, I haven't reviewed the filing itself, and the resolution doesn't explain what drives the loss figure. The resolution is symbolic; it can't force anything. Anthropic hasn't commented in the reports we reviewed.

Students Are Noticing

The National Student Clearinghouse reports computer and information science enrollment fell 12.4% at four-year colleges, 12.3% at two-year colleges, and 10.4% in graduate programs this fall. Over two years, undergraduate CS enrollment is down 18.4% and graduate CS down 27.1%. Coverage links this to AI shrinking entry-level coding jobs; the data itself shows the decline, not the cause.

For freelancers and software builders navigating this transition, utilizing our Freelancer Prompt Library can help automate repetitive scaffolding while keeping strategic system design in human hands.

We Defer to AI More Than We Realize

Anthropic study. Researchers gave 50 professional analysts a causal-inference task while an AI agent (Claude Opus 4.8) nudged four analysis choices in randomly assigned directions. Analysts adopted the agent's lean 86% of the time and reused its arguments in 77% of decisions, but credited the AI as a source in only 15%. The authors read this as substantial unrecognized deference on real judgment calls.

Power, and a Probe

The Department of Energy committed a $5 billion loan to Constellation Energy to add about 1.1 gigawatts through reactor upgrades, with the output already sold to Google, Amazon, and Meta. The Sierra Club projects that planned AI projects could lift power-sector emissions to 2.1 billion tons by 2030, erasing 15 years of grid progress.

Separately, the Justice Department is reportedly examining whether Nvidia's $12.9 billion Hugging Face purchase would give it a “foundational point of distribution” for AI; Nvidia says the platform will stay neutral.

What to Watch

  • Coming weeks: Independent benchmarks of Gemini 4 Carbon.
  • Washington: Whether the no-bailout resolution gains cosponsors.
  • Campuses: Spring enrollment numbers for computer science.

Our Take (Opinion)

The capex figures and the enrollment figures are two bets on the same question: will AI do the work? The industry is spending hundreds of billions that it will. Students are quietly hedging.