🧠 We asked Fable 5.1 to build WhipFlow → it did

Anthropic shipped Claude Fable 5.1 on Sep 1. It’s the public Mythos-class model on Claude.ai, Claude Code, and the API. Mythos 5.1 is the same model with lighter safeguards, and it stays behind Anthropic’s trusted access programs for cyber and life sciences, there is no open Mythos signup.

We tried it on day zero. Script writing for ai.samaritan barely changed. Judgement did. We asked it with /goal to build a WisprFlow clone so we wouldn’t keep paying for a dictation sub. It figured out Apple speech and Apple Intelligence were already on the Mac, and built the clone for free. It was not particularly faster than other Anthropic models. On an internal job board we were also building, GPT 5.6 Sol finished first.

If your next job is a messy multi-step build where “what should we use that’s already free on this machine” matters more than raw speed, switch Claude Code or Claude.ai to Fable 5.1 and give it that kind of goal. Cache reads are 75% cheaper than Fable 5, so typical token workloads land about 25% lighter.

World Labs launched Atlas, an omni world model that works across text, images, video, and 3D. It can generate camera-controlled video, reconstruct a scene from a few photos, and help with Real-to-Sim style robotics work. Future versions of Marble are meant to run on it.

What you can do today is request early access. The page ends in a form for select partners, not a public playground. If you’re building anything that needs controllable 3D or spatial video, put your name in the queue. All we can do is wait (for now).

Runway introduced Solaris on Aug 31. It’s their first Interface World Model, instead of coded screens, the model renders the interface frame-by-frame as you click, drag, or talk to it. Think: drag ingredients into a salad and watch them combine. Rearrange furniture and see reflections shift. Try on clothes against your body.

The scene itself is the UI. No intermediate layer. No translation loss. No waiting for updates. What you can do now: Request early access. Launching with key partners first.

Google Pics is prompt-to-design for images inside Docs and Slides, with Drive later. It is rolling out to Google AI Pro and Ultra subscribers and most Workspace business customers. Open pics.new, sign in, and generate one social graphic or slide asset. Cold free access is not the path, you need a paid Google AI or qualifying Workspace plan.

Clipto is a local-first Memory for videos, meetings, voice notes, images, and docs on your computer. Search a moment by what was said and land on the timestamp. Hook the same Memory into Claude, ChatGPT, or Cursor over MCP. Needs Apple silicon Mac with 16GB+ or Windows with 12GB+. Start with one messy folder of calls or footage and ask for a decision you already made.

Ben Lang shared how different teams at SpaceXAI are actually using Grok Bot: Engineering, Support, Design, GTM, Product Management, and Field Engineering.

Instead of generic prompts, they're using chief-of-staff routing, pinning their most-used agents, keeping living lists of what's open, and teaching tasks once. Open the thread, pick one habit, and apply it to the agent roster you already have.