🧠 Anthropic's Back. The Sonnet's Better

Remember when Anthropic pulled its two most advanced models off the table three weeks ago? The Commerce Department had ordered it to cut off access for any foreign national anywhere in the world, citing national security concerns tied to potential misuse. Anthropic complied within days and went quiet.

That standoff is over. The government lifted the export controls, and Anthropic says access starts rolling back starting today. Fable 5 returns for global users on Claude.ai, the Claude Platform, and Claude Code, now with an added safeguard aimed at the specific jailbreak risk officials flagged. Mythos 5 is also coming back, though for now it's limited to a smaller group of government-approved organizations rather than everyone.

If you had workflows built around either model before the freeze, this is your green light to pick them back up.

Google added a third video format to NotebookLM, its tool for turning uploaded documents into study material. Alongside the existing narrated-slide overviews and the longer cinematic ones, you now get Short Video Overviews: vertical, 60-second clips that walk through a single concept from your sources with narration and animation.

The idea is simple. Instead of rereading a dense PDF, you get something built for the way people already consume video. It's rolling out now to Google AI Ultra and Pro subscribers on mobile and web, with free access coming later. English only for now, with more languages expected to follow.

Anthropic also shipped a new Sonnet, and it's a bigger jump than usual. Sonnet 5 can plan, use tools like browsers and terminals, and run autonomously at a level that used to require the larger, pricier models. Early testers say it finishes complex, multi-step tasks that older Sonnet versions would stall out on, and it checks its own output without being asked to.

Performance-wise, it's now sitting close to Opus 4.8 on reasoning, coding, and tool use, at a fraction of the cost. It's live everywhere today, it's the new default on Free and Pro, and it's available on Max, Team, and Enterprise too. Pricing starts at $2 per million input tokens and $10 per million output tokens through August 31, after which it moves to $3 and $15.

If you've been leaning on Opus for anything agentic, it might be worth testing Sonnet 5 first now that the gap has closed this much.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.

Quick context if you're new here. PAI Labs is our paid track where we hand you full AI workflows you can copy and run on your own business. This week's lab takes on the thing every founder says they'll get to and never does: personalizing outreach past the first ten names on the list.

You know how this goes. The first few messages you send are genuinely tailored. By lead thirty you're copy-pasting the same template with a name swapped in, and the reply rate shows it.

The setup works like a research assistant with strict rules. You paste a lead's LinkedIn profile (About, Experience, recent posts) into Claude along with a one-time sender brief covering who you are, what you offer, and what a yes looks like. Claude finds the single strongest, verifiable trigger on that profile, matches it to the opener that performs best for it, and drafts a 3-part warm LinkedIn sequence plus a cold email backup, enforcing character limits so nothing gets cut off mid-send. If the profile is thin, it says so instead of inventing a reason to reach out. Add Apify's LinkedIn scrapers and Claude pulls the profile data straight from a URL, no copy-pasting required.

Read the full walkthrough here.

Google dropped a new image model built purely for speed. It generates a 1K-resolution image in about four seconds for $0.034, making it the cheapest, fastest option in the Nano Banana lineup so far. It's live now in Google AI Studio and the Gemini API, and it's already rolling out across consumer surfaces like the Gemini app, Search's AI Mode, Photos, and NotebookLM.

If your work involves generating a lot of visuals quickly, think A/B testing ad creative or drafting dozens of product mockups, this is the model built for that.

Anthropic launched a dedicated app for researchers. It connects to more than 60 scientific databases, manages compute environments on demand, and traces every artifact it produces back to the code that generated it, so results can actually be reproduced later. It's in beta now on macOS and Linux.

Not the audience for most of us, but worth knowing about if research or biotech touches any part of your world.

You can now use Google's new Nano Banana 2 Lite and Gemini Omni Flash pair to create high-quality videos using simple prompting:

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.