September 29 - October 4, 2026
The catch-up edition. Six stories in plain language, each with a link if you want more.
In this edition
Tue, Sept 29
Anthropic's new everyday model is faster and cheaper
Anthropic released Claude Sonnet 5.5, the mid-size model most people use for daily work like fixing code and creating documents, slides and spreadsheets. Anthropic says it runs more than 30% faster and costs up to 30% less for most work than the version before it, at the same per-token price. Benchmark scores are Anthropic's own.
Why it matters: Faster and cheaper everyday AI lowers the cost of using it across a business.
In plain English: Anthropic released a faster, cheaper version of the AI model most people use day to day.
Wed, Sept 30
OpenAI introduces assistants that keep working while you do other things
At its developer conference, OpenAI announced "dots," always-on AI agents that run on their own cloud computers and keep working toward a goal over time. They are available to Pro and Business Premium users in eligible markets. OpenAI also launched GPT-6.1 Sol, which it says nearly matches its top model, GPT-6 Astra, on coding and computer tasks at one-fifth of the token price.
Why it matters: Cheaper, steadier AI is what lets more businesses use it for real work. Prices and performance are OpenAI's own figures.
In plain English: OpenAI announced AI assistants that keep working in the background, and a cheaper model that it says performs close to its best one.
Thu, Oct 1
Google announces Gemini 4 Argon, built for long, complex jobs
Google announced Gemini 4 Argon, designed for long, multi-step work in software, legal and finance work, and cybersecurity defense. For now it is rolling out to a group of trusted cyber defenders first, so they can use it to find and fix software weaknesses before wider access opens. Google says it will expand access in phases.
Why it matters: A defender-first release shows how the leading labs are staging their strongest models.
In plain English: Google has a new model for long, complex work. Security defenders get it first, and wider access follows in phases.
Fri, Oct 2
Cloudflare releases free "decision" models anyone can download
Cloudflare, which runs a large part of the internet's infrastructure, released Clef and Clef-flash, open models that answer structured questions such as yes or no, multiple choice and rankings, rather than writing text. They work like TypeSafe's Jev, covered in Volume 1, and can run on a company's own machines. Performance comparisons are Cloudflare's own.
Why it matters: Decision-style AI is quickly becoming a category, and open versions give companies a way to run it themselves.
In plain English: Cloudflare released free AI models that give structured answers, such as yes or no. Companies can run them on their own machines.
Sat, Oct 3
A small free AI writes cited science reports in under a minute
The Allen Institute for AI (Ai2) open-sourced AstaBrief, a compact model that turns a research question and a set of papers into a cited report. Ai2 says its fast mode averages 51 seconds per report, against about 3 minutes for its Claude-powered mode. Researchers can download it and run it on their own computers.
Why it matters: Open models that run locally matter for researchers whose questions are sensitive.
In plain English: A small, free AI can turn research papers into a cited report in about a minute. Researchers can run it on their own computers.
Sun, Oct 4
Almost half of people thought this AI on a video call was human
Tavus introduced Griffin, a video AI that talks live and picks up on pauses and expressions. In a live study, 48% of participants thought they were talking to a real person; earlier systems, including Tavus's own, topped out at 2%. The result is Tavus's own study.
Why it matters: Live video AI is getting close to natural conversation, which makes clear labeling of what is AI more important.
In plain English: A video AI was convincing enough that nearly half of test participants thought it was a person. This is Tavus's own study.
4 pages. Each story has a link if you want more.
Putting the human back in AI.