Antigravity's computer use has also been massively upgraded with Gemini 3 Flash. It is both faster and better at doing long agentic tasks using the browser. Here's Antigravity doing deep research on the Pareto frontier of models and writing code to visualize the result.

I spent many (unsuccessful) hours trying this just 6 months ago. Feeling the AGY.

Shout out to @OriolVinyalsML for the idea!

@OriolVinyalsML Yes and lots of work from research and product to make this happen, special shout out to @mooohul and team for the deep work for making the experience smooth.

If Antigravity were open source, imagine what we users could build - with and for it. Using Gemini 3 Flash and Pro, we could enhance the app itself and make Antigravity the best IDE and agent orchestration platform.

thank u for ur work on antigravity. the last time an ai tool changed my life like this was when openai first introduced codex in 2021

some feedback: I really wish the computer use screen was integrated into Antigravity, like in AI Studio Build Mode. Maybe it could go in a separate tab within AG (where the replay currently shows) or inside the Agent Builder. This could make the testing process (by me)more smooth

Tried it today and it's amazing. Gemini 3 Flash is an amazing model. It beats Sonnet on SWE-Bench Verified. It's also faster.

Today I actually used it to migrate my old WordPress site over to vite, I was busy to start copying stuff so I gave it a link and it went studied the page live scrolled picked up the content wrote it down, I did a few tweaks on the content copy It went ahead and did the design

@demishassabis Man that IDE laggy as fuck and I got a really good laptop too Long threads More than one window And its unusable

Amazing work @_mohansolo

No offense but the web fetching functionality in Antigravity is worse than both Github Copilot and Cursor. It's sitting there scrolling a browser for 5 mins struggling to get over EU cookies and lazy loading while I'm staring right at the thing it requires...

A little bit of feedback: on an M1 Pro MacBook Pro with 32GB of RAM, Antigravity chugs so badly *when* the agent is running, it is literally impossible to click around/scroll around in the IDE or in the agent output. I hope this can be fixed.

Need to try it out. Gemini 3 pro was slow and disappointing when compared to Clade code.

The version control inside the agentic chat window is just deleting existing code files instead of reverting the files back to pre-prompt state This happens whenever I click "undo changes up to this point" on previous prompts Fix it please

Sounds like Antigravity's cooking up some serious wizardry 😂

Impressive upgrade! Excited to see what Antigravity can achieve now.

Why does it always break 💔

Impressive upgrade, making research and coding more efficient!

Is there a way in Antigravity to see if you're approaching usage limits for various models like 3 Flash, 3 Pro, and Opus 4.5?

leave eyething aside please please imrpove the stability of ur ide its shit agent crashes very frequently

my agent keeps terminating and deleting the whole conversation in process !!

antigravity is underrated

What is the best way I can leverage Claude Skills on Antigravity? is there a way?

Seeing that for the first reminded me of the days when VNC and similar programs first arrived, seeing someone, or being that person driving a desktop from far away (business, not some scammer lmao). Seeing AI do it is another level.

Maybe I will try out Antigravity, but claude code is working so great for me that I don't have need to test other coding tools...

man antigravity kills my m1 pro macbook with 32GB of ram

I have had the most frustrating experience with this tool Tip : Open antigravity and open claude code , give them the same task and see side by side comparison By the time antigravity figured env setup , claude had mvp version 1 ready It's not even close

i wish it could be disabled, i know settings for that exist, but neither flash, neither pro, not even opus seem to respect them and wants to turn on the agent browser often

Computer use is finally becoming a real product surface: plan → act → verify → iterate. The differentiator won’t be “agentic”—it’ll be reliability under messy web constraints: auth, popups, latency, and graceful recovery when the world changes.

@antigravity Always freezes right after I open a project, None of the updates have fixed it. - Reinstall did not work - Disconnecting VPN did not work - Deleting Cache did not work OS - Win 11 This does not look like a release ready product from google - any ETA on the fixes

Can I use this solely as a computer use model? Like replacing Comet from perplexity

Yo brotha, give me a job at google. I am pretty good at coding.

ultra fast

Can't wait for the RL to push into pro and make an absolute beast

I need to download chrome and give this try, forgot how to add a different chromium browser.