No. 1

The Daily Diff

4 min read

Google ships Gemini 4 Neon, its fourth model in seven weeks

A grey microchip with rows of pins along its left and top edges and a square die at its centre, standing in for an AI model.
Illustration: The Daily Diff

Google has released Gemini 4 Neon, a large language modelAn AI system trained on large amounts of text that reads and writes language.. That is the kind of AI that reads and writes text. Google says it is built for software development, knowledge work and science. It accepts text, images, video, audio and PDF files, with a context windowHow much text, images or other input a model can consider at one time. of one million tokens. The context window is how much input the model can consider at once.

Some results come from outside testers. Artificial Analysis gives Neon 61 points and ranks it third overall, behind two rival models that Google left out of its own comparison table. Google says Neon is nearly twice as fast as its sibling, Gemini 4 Carbon.

Several things are still uncertain. Many benchmark scores come from Google's own tests. Neon's list price is higher than some earlier Gemini 4 models. Real-world results may differ from the tests.

Google is rolling Neon out in the Gemini API, Google AI Studio and the Gemini app for paid subscribers. Developers will want to test it on their own work before switching.

Why it mattersTeams weighing AI tools for coding and office work now have another option, priced above earlier Gemini 4 models.

Page 2The Daily Diff · Thursday, November 19, 2026

AI

Models, research, tools, policy
A brass-style bell with a loop crown, flared mouth and hanging clapper, standing on a soft shadow.
Illustration: The Daily Diff

Anthropic prices its IPO at $83 a share

Anthropic set the price of its initial public offering at $83.00 per share. The offering is 980 million shares, most sold by Anthropic. The shares are due to start trading on Nasdaq on 19 November under the symbol ANT.

Why it mattersAnthropic's share sale will show how investors value a leading AI lab and may shape other AI company listings.

AI answers have cut traffic to a Berlin travel guide sharply

Nicolas Bouliane, who runs the All About Berlin guide, says Google's AI Overviews have cut its traffic by 70% to 75%. He says he may need another way to pay his rent.

Why it mattersPublishers that depend on search traffic may lose readers as AI answers take their place, which affects what gets funded.

Genie 4 world model arrives for Google AI Ultra subscribers

Google is rolling out Genie 4, its newest world modelAn AI system that generates an interactive, simulated environment you can move through., to eligible Google AI Ultra subscribers. Users can change their generated worlds in real time with text prompts, and add native audio.

Why it mattersSubscribers can build, share and remix interactive worlds in a browser through a new community gallery.

KPMG's AI agents out-questioned its audit team in a trial

KPMG UK tested more than 100 AI agents on the accounts of an unnamed FTSE 250 company. Sometimes they dug deeper and asked better questions than the human audit team. The results were not used to sign off accounts.

Why it mattersAudit firms may face pressure over fees and staffing if AI agents can do much of the detailed checking work.

Page 3The Daily Diff · Thursday, November 19, 2026

Tech & Industry

Companies, products, security, hardware
A closed steel padlock with an arched shackle and a keyhole in its body, drawn in shades of gray on a bare page.
Illustration: The Daily Diff

Ruby 4.0.10, 3.4.13 and 3.3.14 fix seven security flaws

Ruby has released fixes for seven vulnerabilities in its regular expression engine and string case conversion. Some have known proof-of-concept code that shows remote code executionA flaw that lets an attacker run their own code on someone else's computer over a network., so the Ruby team urges upgrading.

Why it mattersAny app running an affected Ruby version should upgrade now, since some flaws can allow remote code executionA flaw that lets an attacker run their own code on someone else's computer over a network..

Two critical NGINX flaws allow unauthenticated code execution

Researchers found two critical flaws, CVE-2026-123633 and CVE-2026-123007, in the NGINX web server. Both can let attackers run code on a server without logging in, and both need HTTP/2 enabled.

Why it mattersSites that use NGINX as their front door should check their version and install the 18 November security release.

How to tell a cyberattack from an ordinary outage

After a run of outages in the US, WIRED sets out how to check whether one is a cyberattack. Security researchers say most outages have ordinary causes, such as a bad update or a configuration change.

Why it mattersBefore blaming hackers, check official status pages, since most big outages turn out to be self-inflicted.

Microsoft confirms apps fail to install after November hotpatch

Microsoft says some Office and third-party apps may fail to install or launch on Windows 11 devices that took the November 2026 hotpatchA Windows update installed in memory without a restart, typically to deliver security fixes.. It has shared a registry workaround for IT admins until a fix is available.

Why it mattersIT teams using Autopatch hotpatches should check affected devices and consider the workaround.

Page 4The Daily Diff · Thursday, November 19, 2026

Repos of the Week

Week of Monday, November 16, 2026
Page 5The Daily Diff · Thursday, November 19, 2026

Quick Bits

The rest, in one line each

You're caught up.

That's the news. You may now return to pretending the tests are still running.

All past papers · RSS