One AI training opt-out for Google and OpenAI, plus Claude Ultra

Anthropic renames Claude Max 20x to Ultra and drops the five-hour limit

Anthropic’s top personal plan is now called Claude Ultra. It is the old Max 20x plan, still US$200 a month, and current subscribers move over automatically. Verified Ultra accounts no longer have a five-hour usage limit.

Ultra gives you 10 times the Pro plan’s weekly usage. A weekly limit still applies across all models, and Fable has its own weekly limit. Verification is automatic. If your Settings, Usage page shows no “Current session” bar, you are verified.

Why it matters Long coding or Cowork sessions no longer stop halfway through the afternoon. If you lean on Claude Code all day, the weekly cap is now the only number to watch.

I think the five-hour wall was the most annoying limit on any AI plan. I’m glad to see it go.

Read the plan details on the Claude Help Center

Google’s Gary Illyes says AI has made fake Googlebots worse

Gary Illyes from Google says bots that pretend to be Googlebot are a bigger problem now, and AI is the reason. Companies want content to train models, so some scrape the web while using a fake Googlebot name. His own test traps caught “Googlebot” visits from IP addresses that have nothing to do with Google.

He also called robots.txt a suggestion, not a lock. He is looking at signed bot requests, called Web Bot Auth, as a better way to prove who a bot is.

Why it matters A user agent that says Googlebot proves nothing. Check Google’s published IP ranges or do a reverse DNS lookup before you trust it. A firewall from Cloudflare or your host can block the fakes for you.

Read the full story on Search Engine Roundtable

Vercel Data Cache becomes Runtime Cache, and reads and writes get billed

Vercel’s Data Cache is leaving beta and merging into Runtime Cache. Next.js fetch caching and unstable_cache move over with no code changes. Pro projects get 8 times more cache storage than before.

The catch is the bill. Cache reads and writes that used to be free will now be charged. Rates start at US$0.40 per million reads and US$4 per million writes. Hobby teams get 1 million reads and 200,000 writes free each month and move on 29 November. Pro teams will get a date by email.

Why it matters A busy Next.js site can make a lot of cache calls. Check your Data Cache usage in Observability now, before the first new bill arrives.

Read the changelog on Vercel

DoorDash says AI shopping agents sent only a few hundred orders

DoorDash says third-party AI agents have placed only “several hundred” orders so far. That is tiny next to the roughly 3 million deliveries it handles a day. CEO Tony Xu said most people still order in DoorDash’s own app.

DoorDash still wants to learn from agent makers, and it may advertise inside their agents later. Amazon blocks personal shopping agents, while Walmart welcomes them.

Why it matters AI shopping agents get a lot of hype, but real orders are still small. My advice for shop owners: make your products easy for agents to read, and keep your own checkout great. That is where the sales are today.

Read the full story on TechCrunch

Cloudflare’s new DNS page is now the only one, and it can change record types

Cloudflare has retired its classic DNS records page. The new DNS records experience, first shown in May, is now the default for every plan. You can now change a record’s type in place, for example from an A record to a CNAME.

Why it matters If you manage DNS for clients, the page will look different next time you log in. Changing a record type no longer means deleting and re-adding it, which removes one easy way to break a site.

Read the changelog on Cloudflare

HTTP/3 made Delivery Hero’s app 17% faster, with Cloudflare’s help

Delivery Hero moved about 95% of its app users to HTTP/3 and cut home screen load time by 17%. Engineer Elvis Chidera shared the numbers at the P99 conference. Devices on networks that block UDP fall back to HTTP/2.

The secure handshake got 51% faster. In Pakistan, where a handshake could take nearly a second, response time was halved. Some pages on patchy networks are still slow, and the team is working with Cloudflare on a fix.

Why it matters HTTP/3 helps most where mobile networks are slow. If your site is on Cloudflare, it is one switch in the Network settings. It is one of the first things I would check.

Read the full story on The Register

Tokio’s creator launches Topcoat for building web apps in Rust

Carl Lerche, who created the popular Tokio library for Rust, has shown Topcoat. It is a framework for building web apps in Rust. Most logic runs on the server, and when you click a button, only that part of the page comes back as fresh HTML.

Small bits of browser logic can be written in Rust and compiled to JavaScript. Data can stream into placeholders, so a slow query doesn’t block the whole page. Lerche says it suits smaller, database-driven apps best.

Why it matters It is a fresh take on fast, server-rendered pages. If you already write Rust, it is worth a weekend test.

Read the full story on DevClass

Apple tells stores how to fix the iPhone Duo’s cracked inner screen

Apple has sent a memo to Apple Stores and repair partners about inner screen cracks on the iPhone Duo. Technicians should inspect the phone, then repair or replace it under normal rules. Apple says only a small number of phones are affected, and there is no recall.

Apple says the fault tends to show up soon after you start using the phone. iFixit thinks a faulty hinge is the likely cause. Apple has not set up a public repair program yet.

Why it matters If your new iPhone Duo shows a crack on the inner screen, go to Apple support early. It also says something about folding phones: test your site on them, but don’t build for them first.

Read the full story on MacRumors