Opus 5.5 killed the game. We had it and GPT-6 Sol both draw Dario in Canva with ego (lite). Claude Opus 5.5: 1h 23m, $6.31 GPT-6 Sol: 26m, $14.97 Opus took 3x longer, cost less than half, and drew a much better Dario.

This week, we shipped ego (lite) Windows beta. It’s now open to everyone in our Discord. And honestly, that’s kind of how we like building now. Early users test things, tell us what’s broken, push for features, and help us decide what deserves attention next. If you want early builds and a closer seat to what we’re building: come hang out →

You asked how Jev could work with ego lite. So we gave it a fun job first: judge the X feed in real time. ego scrolls. Jev decides: slop or not. repeat. Fast decisions are a lot more fun when you can actually watch them happen. (toy experiment — the slop detector itself is definitely not perfect yet 💀)

ego lite + Jev + DeepSeek Flash = stupid fast. ⚡ 3.71s for 20 Amazon product decisions. GPT-5.6 Sol: 54.45s Same task. Same result. 10/10 on both. Less reasoning. Faster decisions. 1× speed.

Okay, GPT-6 Astra is getting scary good. One prompt: research game keywords → pick an opportunity → build the site. 15m27s. $5.09. No follow-ups. It researched, made the call, built the site, tested it, and saved the outputs by itself. We just gave it ego (lite) and got out of the way.

Cheap models make the browser tool your biggest lever now. Here's what that looks like in practice. We ran the same hotel search on DeepSeek V4.1 Flash twice. Playwright took 9m23s and $0.20. ego (lite) finished in 8m17s for roughly half the cost.

ego (lite) is coming to Windows. We're keeping spots limited and prioritizing power users who want to help shape the Windows experience from day one. Want in? Apply here →

GPT can now learn skills by watching YouTube tutorials with ego (lite). We had GPT-5.6 Sol use ego (lite) to watch a YouTube tutorial at 1.5x. Then it switched Spaces and recreated the quiz workflow in Google Forms from memory. 36m50s. 64 ego-browser calls. Totally executable.

This is not the Wispr Flow website. GPT-6 Astra used ego (lite) to inspect the real one, then rebuilt it almost pixel for pixel. 28m33s. $8.02. The browser was the design file.

GPT Image 2.5 turned studio-quality product shots into a prompt. With ego (lite), GPT-6 Astra sketched the idea in ChatGPT, generated the final image, and uploaded it straight to Shopify. 10m16s end to end. $5.4 in model cost. GPT Image 2.5 made the asset. ego (lite) put it to work.

Major engine upgrade just dropped. 20% fewer tokens per task. Faster execution. Higher success rate. Profile-aware Spaces: agents can now switch between browser profiles and spin up Spaces per profile. No more session bleed. Update now and try it.

$0.049. DeepSeek V4.1 Flash used ego (lite) to shop Amazon for a MacBook desk setup, all under $100. V4.1 Flash was pushing 295 tok/s. ego (lite) kept the clicks moving. The cart had a productive five minutes.

GPT-6 Astra cleared all 48 levels of "I'm Not a Robot" with ego (lite). With the built-in browser, it got stuck at level 15. Same model. Same game. Apparently, the browser had to prove itself too.

You might not have noticed, but someone fixed a years-old issue with Chrome on Mac. Window resizing, tab switching, things you do every day. The animation frame rate has long been less than half of your display's refresh rate. You felt that "not quite smooth" feeling, but probably never thought it shouldn't be that way. The video below shows it. As of Chrome 147, it's fixed. You don't even need to be one of our users, just update Chrome to the latest version and you get it. The person behind this is our client lead @liuwd_citro. He's landed 300+ changes in Chromium, focused on the small details that make Chrome faster and smoother, most people don't notice this work, but everyone benefits from it every day. Why do we do this? Cause we stand on the shoulders of countless open source projects. Building something genuinely exciting isn't possible without that foundation, so we wanna give some back. And in a moment when AI slop is flooding the world, we still believe good products live in the details. What's been bugging you about Chrome for a long time? Drop it in the replies. We will read every one of them. Where we can, we'll patch it. And we'll prioritize giving you early access to what we're building.

Dear developers (and anyone who's interested in the future interface of web): our boy @imCGQAQ just got shape() accepted as a valid value for shape-outside, which means text wraps around any curve you draw, on a left-floated figure, in one line of CSS, for every Chrome user on Earth. Chrome 149 will ship this feature next week. shape-outside has been in the spec since 2014. for eleven years it accepted exactly five values: circle, ellipse, polygon, inset, image. want a Bézier? approximate it with 40 hand-counted vertices and live with the jagged edges. that compromise is over. and yes — overlaps with @_chenglou's pretext. the difference: pretext is a JS text engine in userland, you build any layout on top. ours is one CSS property in the browser itself, applied to you. opposite ends of the stack. 3 demos. each is a real CSS file, no JS layout code. ↓