Gemini 4 Carbon - A Mechanical flower! This took ~1hr 43mins fully made in three.js, I love the small bee it added which wasn't in the prompt but I did say it to be creative and didn't limit it with constraints.. A lot of improvements over Argon tbh.. I'll share my observations once i have some more outputs.. I've got one more and still trying to get more.. stay tuned!

Gemini 4 Carbon's checkpoint got updated! The last one was very slow so let's see this one got any better in speed.. It's a pain to find it on Arena but sure :)

Claude duet? Powered by Opus 5.5? Count me in.. Yup it's over for Dots.. the only competition it'll have will be from Grok bot.. since Elon announced that they'll also use Opus 5.5 in Grok bot.. and it has a larger ecosystem built around it.. Sad that it's US only for noww ;)

Gemini 4 Carbon - A Mechanical bee! Did my mechanical Bee test with Carbon.. It's actually better than Argon on the overall polish and detailing.. It nailed the bee's waggle dance which they do in a figure 8 pattern and the wing movement! See how Argon did in the quoted post!

So I was trying to get some outputs from Gemini 4 Carbon in arena on one of my Mechanical Bee tests.. I got Mimo-v2.6-pro there and lmao this was so funnyy 🤣 I couldn't stop laughing when i saw what it did with the Bee's flight.. bro broke my bee.. wait till I click land😭

Yup.. they nerfed Opus 5.5!! So as usual, I let it run on one of my projects overnight.. a crucial feature and some bugs to sort through. I'd already seen a lot of people saying it got nerfed, so I was pretty skeptical. Woke up this morning and went through its work. Saw some of the stuff it did and a few decisions it took.. oh my.. Told it to run some tests and stepped away. Came back to every app window transparent and unresponsive. For the first time ever, I saw the "Microsoft Windows isn't responding" error.. Restarted my PC and asked Opus what happened. Bro ran some unbound script that tanked my whole PC.. Today was probably the only time I had to say these two sentences to Opus: ~ Leave it, I'll handle it myself.. ~ bruh tf did uh do.. Usage is also draining faster than it was two days ago.. not gud anthropic not gud..

Seems like Gemini 4 Carbon is coming pretty soon with meaningful upgrades over Gemini 4 Argon.. We might see it in the next couple weeks..!

Good news.. the AGY folks are already discussing letting us set our own compaction threshold in Antigravity! Currently agy compacts at ~256K by default, and it's based on evals around speed and performance.. This is why I love giving feedback to this team.. I remember my first review feedback was of the Antigravity IDE on Aug 9 back then i had a couple followers and no one knew me.. And they still responded and most of it was acted on.. since then every feedback has been taken seriously and a lot of the community feedback has actually shipped ! Thanks Jack and the AGY team really hoping that this ships soon as it will help a lot!

It seems like Antigravity is suffering an outage as we speak.. I'm also getting Model Unavailable, Retrying attempts and slow model outputs! Hopefully team is working on it!

So I got some DMs and questions about the unusual workflow.. using Argon in the Agy 2.0 app to manage multiple agy CLI workers instead of just letting agy spawn subagents. Here's why This is what the setup looks like: Agy 2.0 app (Argon, primary orchestrator) ├─ Project A → agy CLI worker │ ├─ subagent │ └─ subagent ├─ Project B → agy CLI worker │ └─ subagen (if needed) └─ Project C → agy CLI worker ├─ subagent ├─ subagent └─ subagent ~ The Agy 2.0 app is the one place I look at as it has the context of every project.. ~ Each project gets its own agy CLI instance working and reporting back up to it.. ~ Each CLI spawns its own subagents when it needs to or when told to depending on the proj.. Why I went with it: ~ Range. I could test Argon across a lot of projects at once from one thread ~ Managing. I was also able to test how good Argon is at managing multiple workers at the same time and holding the context of each instance and the project it's on ~ Going back and forth. One agent knew every project's decisions, which gates applied where and how to deal with the recurring errors. Sign-in expired partway through 41 of my 167 CLI runs.. most of those were stale auth errors that didn't actually mean anything, and it just restarted the worker after a minute or two ~ My machine. Multiple projects running their own tests, Chrome instances and benchmarks at the same time could OOM my PC. It handled that by giving each project's workers their own time slots for the heavy stuff ~ I'm not a CLI person... :) I'd never used agy CLI before and CLIs usually don't go well with me. This way I didn't have to sit and go through it, and I still got to test it properly.. there's a feedback report on it dropping soon So yeah.. I needed to run multiple things in parallel, build projects, collect data, run different benchmarks and still see everything in one place. The thread and every number in it wouldn't have been possible without this setup.. One alternative i considered was @theo 's T3 Code.. the orch V2 is goated but the agy implementation is not that smooth it might be due to how i wanted to work or something else.. i did try it but then went back to Agy..

I didn't waste anything lmao? I have built multiple projects using both across that time period.. I don't know how well you know me and if you have seen my previous posts.. Everyone knows Opus 5.5 is the best model rn without question.. the thread & the analysis isn't about helping me in deciding the best model.. I'm simply evaluating and sharing my experience with Argon and how should you actually use it to get the best results.. and if it can help someone in any way I'm very happy to spend some time to form a report and share my workflow..