In partnership with
Hello friends!
Welcome to this week's Sloth Bytes. Two frontier models dropped in three days, and your IDE got dragged into a lab beef.
For the people with low attention spans:
GPT-6 Astra and Claude Fable 5.1 are out
OpenClaw 2.0 is the biggest update they have ever shipped, and they did it by accident
OpenAI is cutting Cursor off after the SpaceX deal. Shutoff happens November 12.
htmx 4.0 and Vitest 5.0 also came out
Latest Posts
If you're new, or not yet a subscriber, or just plain missed it, here are some of our recent editions:
1,000+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
1,000+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
GPT-6 Astra vs Fable 5.1
GPT-6 Astra just came out (OpenAI's announcement)
Price: $10 in / $50 out per million tokens
Best for: computer use tasks. QA testing, automations, anything where it has to actually click through a real interface instead of describing one.
Also strong at: general use and knowledge work, plus long-running tasks that have to hold together over many steps
Bonus: the team is giving out resets each day you don’t have access to it.
Claude Fable 5.1 also came out (Anthropic's announcement)
Price: same $10 / $50 as Fable 5 and Astra
Best for: thinking. Also more ambitious coding work, the kind where it needs to hold a whole architecture in its head instead of patching one file.
The real change: Cache reads cost 75% less, making it $0.25 per million tokens. Cache reads are what you pay every time the model re-reads context it has already seen, which is basically every single turn of an agent loop.
What that saves: Anthropic says about 25% on typical workloads, up to about 45% on agentic ones
So which one do you actually use?
Computer use, QA testing, or a long-running job: Astra.
Ambitious coding, frontend work, or more readable code: Fable.
The most usage: Astra with Codex. Just because of how many resets the team gives you.
You just want the cheapest thing that works: neither. Put a small model on the boring stuff and stop paying frontier prices to rename variables.
And take every benchmark with a fistful of salt.
On the public DeepSWE leaderboard, Gemini 3.8 Flash ties Astra at 74%.
Yes. A Flash model tied with the "welcome to the AGI era" model…
So please don’t rely fully on these benchmarks. These models optimize to do well on them.
Oh right, I guess Google released Gemini 3.8 Flash. It’s something…
Cute model, Google, but please tell me you’re cooking something crazy with Gemini 4.
OpenClaw 2.0 is out (if you care?)
So apparently OpenClaw is still a thing. I thought it died…
Quick refresher for you: OpenClaw is an open source AI assistant. Basically it’s meant to be Jarvis where it can handle tasks for you without you having to ask.
A version 2.0 wasn’t supposed to happen. The announcement post is literally titled "OpenClaw 2.0, Accidentally."
Their original goal was to simplify installation and rebuild the browser app, but that got more complicated fast…
After 933 contributors, 569 first-timers, and 16,000+ pull requests, they ended up with a version 2.0. Imagine refactoring so hard you ship a major version.
What actually changed:
Installation should be easier: It starts from whatever is already on your machine, your ChatGPT or Claude subscription, your API keys, and your local models.
Also simplified a lot of configuration and moved the rest out of initial setup.
Rebuilt browser app: Now it basically looks like ChatGPT/Codex, which is a massive improvement. The original browser app was disgusting…
Multiplayer cloud sessions: Now multiple people can use the same agent.
Heads up before you upgrade: Sessions and transcripts moved into SQLite. Back up your data first. If you downgrade later, anything created after the migration won't show up in the older version.
RIP Manifest V2 on Chrome
All remaining MV2 extensions are gone from the Chrome Web Store. uBlock Origin included. Copies already on Chrome 138 or earlier stay installed. They get no updates and you cannot reinstall them from the store.
Brave is hosting uBlock Origin, AdGuard, uMatrix, and NoScript on its own backend. If you use Chrome, your options for ad blockers are uBO Lite, Brave, or the Firefox migration you keep putting off.
htmx 4.0 is out and finally uses fetch()
htmx 4.0 is out. Internally it dumped XMLHttpRequest for fetch(), which most of you will never notice, but it makes development easier for them.
Here’s the stuff that actually affects you:
Inheritance is opt-in now. In 2.x, parent attributes like
hx-confirmleaked onto children by default. In 4.x you writehx-confirm:inheritedor they don't inherit.Events got renamed.
htmx:beforeRequestis nowhtmx:before:request. Same pattern for swap/config. If you havehx-onlisteners or custom JS, find-and-replace before you ship.History stopped using localStorage. Back button re-fetches the page (or
[hx-history-elt]) instead of restoring a stale DOM snapshot that third-party JS had already mutated. Want the old cache feeling? There's anhx-history-cacheextension onsessionStorage.Morph swaps are built in, plus a clearer
<hx-partial>tag when one response needs to update multiple targets (think new message + bump the count).Upgrade is still opt-in. npm
lateststays on 2.x until early 2027 so unversioned CDNs don't explode. Pin4.0.0when you're ready. 2.x is supported indefinitely.If you do want to upgrade: run
npx [email protected] upgrade-check -- ./templatesand let it yell at you. Or use their agent upgrade skills.
Vitest 5.0 wants your test suite to be fast
Vitest 5.0 is out and it’s mainly a performance improvement. Not many API changes. This is still a major update, so your suite can break in quieter ways.
Here's what actually affects you:
Important breaking change: Vitest 5 requires Vite >= 6.4.0 and Node.js >= 22.12.0.
It gets faster where suites usually die. Biggest wins on vm pools, Browser Mode, and fat monorepos.
Inline projects that don't change Vite config now share one Vite server.
fsModuleCacheis stable (transformed modules persist across reruns).New
vitest doctortries alternate configs and tells you what would actually speed you up.
clearMocks is on by default. Vitest clears mock call history before every test. Implementations stay.
Order-dependent flaky mocks get louder. Set
clearMocks: falseif you liked the old way.
Unawaited async asserts now fail.
expect(promise).resolves...withoutawaitused to warn and still pass. Now the test fails.Browser Mode locators are strict by default, so
getByText('Item')no longer matchesItem 1by accident.
Useful new toys: Trace View in Browser Mode (replay every click/assert as DOM snapshots),
vi.whenfor per-argument mock behavior, and nested projects inherit the root config by default.The benchmarking API was rewritten (
benchis a fixture insidetest(), not a top-level import).
Before you upgrade: read the migration guide, then
npm i vitest@5. In that order.
You may have to pick a side for AI models
Right now most of us treat the model picker like a buffet. We pick whichever we want depending on the task: OpenAI, Claude, Gemini, Grok, Kimi, etc.
But it looks like this freedom of choice might not last forever.
OpenAI told SpaceX it is winding down OpenAI models inside Cursor. Proposed shutoff November 12. No future models either, Astra included.
Same week, opposite direction: Anthropic's Tom Brown said Cursor has been a trusted partner since Sonnet 3.5 and that they will keep growing compute for Claude inside it.
This is bigger than Cursor:
OpenCode, Zed, Copilot, every editor you like, all of them run on access that somebody else grants and can revoke. Even OpenRouter, whose entire pitch is "use every model from one place," only works for as long as every model provider agrees to let it. Nobody signed anything promising that forever.
We have spent two years assuming the model picker is a menu that only grows, but it looks like that’s starting to end. Soon picking an AI tool will feel more like choosing a side. I hope this doesn’t become the case.
How Cloudflare freed ~100TB of RAM on 1.1.1.1 - five cache-layout changes cut per-entry size in half. Inserts got faster too. This is the fun kind of systems post.
Mastery still comes from doing the reps - Agents can finish the ticket without teaching you anything. A completed task is not a rep. If you are three years in, plausible code may arrive faster than your ability to judge it.
ESLint can autofix you into broken code - two non-overlapping edits are not two compatible edits. elm-review applies one fix, then re-analyzes. ESLint batches.
Svelte Bits - copy-paste animated Svelte 5 components (text, backgrounds, UI). A port of React Bits. Tailwind plus TypeScript. Not another runtime changelog.
Typebase - a type-safe backend you write as TypeScript files inside the app you already have. oRPC, Drizzle, better-auth. MIT.
Schedule-X - a FullCalendar-shaped event calendar for React, Vue, Angular, Svelte, and Preact. Drag, resize, dark mode. MIT core.
Cropper.js - drop-in image cropper with a playground. Use it when you need crop UI, not another unzip helper.
Rslint - ESLint-compatible JS/TS linter, built to be fast. Ships TypeScript-ESLint rules out of the box. Experimental.
That’s all from me!
Have a great week, be safe, make good choices, and have fun coding.
If I made a mistake or you have any questions, feel free to comment below or reply to the email!
See you all next week.
What'd you think of today's email?
Want to advertise in Sloth Bytes?
If your company is interested in reaching an audience of developers and programming enthusiasts, you may want to advertise with us here.
