I open way too many AI tabs every week, and most weeks I close them feeling behind. This one was different. The creator behind this roundup, Matt Wolfe, packed seven days of chaos into one tidy video, and I want to break down the parts that actually matter for you.
Anthropic Shipped Claude Fable 5
Let’s start with the headline. Anthropic shipped two models on June 9th: Claude Fable 5 and Claude Mythos 5. The video host clears up a big myth right away. Fable 5 is a “Mythos class” model that sits a tier above Opus, but it is not the locked-down Mythos 5 everyone was panicking about. The real Mythos 5 stays with a small group of cyber defenders and infrastructure providers.
🔑 Quick Facts
- Fable 5 benchmarks near state-of-the-art across the board
- It can work autonomously longer than any past Claude model
- Pricing is steep: $10 per million input tokens, $50 per million output, double Opus 4.8
- Pro, Max, and Team plans get it only until June 22nd, then it needs extra credits
The Refusal and Downgrade Issue
Here’s the part that stirred up X. The original poster confirmed Fable 5 refused to analyze blood work and quietly downgraded certain prompts, including anything touching frontier LLM development. Anthropic’s own docs admitted those downgrades stayed invisible to users. After the backlash, the company walked some of it back within two hours and apologized. The fix? They’ll still block and downgrade, they’ll just tell you when they do it.
Coding Demos: Where the Real Power Shows
The coding demos are where I got hooked. Wolfe shared how Fable one-shotted a 3D game clone into a single HTML file, then built it into a polished playable game called Cube Basher over a few more prompts. He even had it spin up a YouTube clone with a working recommendation feel, plus a desktop tool that turns any article into animated B-roll using just two prompts. That last one solves a real editing chore, and I think that’s the smartest use here.
🍏 Apple’s WWDC Was the Other Giant
The expert ran through the new Siri AI built with help from Google’s Gemini models. Highlights he flagged:
- Personal context across photos, calendars, and messages
- Visual intelligence through the camera and a dedicated Siri app
- A “describe a shortcut” feature that builds automations from plain language
- Spatial reframe to change a photo’s angle, live now in the iOS 27 beta
One catch the creator noted: none of it launches in the EU at first.
🔍 Google Quietly Had a Monster Week
The person who shared it walked through three drops:
- NotebookLM now runs on Gemini 3.5 with a secure cloud computer and 100+ skills, exporting tons of new file formats
- Gemini 3.5 Live Translate does near real-time voice translation, coming to Google Meet and live in AI Studio
- DiffusionGemma generates text using diffusion tech, trading some smarts for blazing on-device speed
⚡ Rapid Fire from the Roundup
Dario Amodei and Sam Altman both published manifestos about sharing AI’s wealth widely, ChatGPT can now send emails, OpenAI filed for an IPO, SpaceX pulled off a record $75B IPO, Coinbase opened up to AI agents, and Midjourney teased mystery hardware.
The thing I appreciate most is how this savvy professional drinks from the firehose so you don’t have to. Want the full demos, the game you can actually play, and every link? Go watch his breakdown. 🎬