In this head-to-head, Gemini 3.6 Flash and 3.5 Flash demonstrate automated code modernization in Antigravity. Watch 3.6 Flash rebuild legacy software faster, with improved latency and output.👇

Read more about 3.6 Flash:

Learn more about this code migration:

What are you even doing?

We want more than Flash

I tried model 3.6 today; it’s even dumber than 3.5. I gave it a task, it completed only a quarter of it and claimed everything was perfect—it’s the dumbest model imaginable. @google, don't embarrass yourselves.

Gemini models have been trash people have been using antigravity models for claude models lol! Time to add sonnet 5 and opus 4.8 btw

Speed is really nothing if you can’t do things accurately. Other labs are going forward. You are going backwards. You should call it -3.5 flash. Decrement your numbers going forward.

Have a nephew freaking out because antigravity 1.23.2 won't open, trying to help him and many reports of the issue on your forums with no response. Tried to get antigravity 2.1.1 up and running for him and none of the brain/conversation migration tricks are working. Now agents are locking up......

Hurry up and go bankrupt!Garbage stuff,It cannot be used on versions below 2.0

It works super fast, but I see no improvement in quality over 2.5

Antigravity is so back #

@antigravity magic 🪄

On prompt and 5 min later, 76% consumed, very fast and efficient.

3.6 flash is really fast

honestly why do they keep naming updates after magic when we're hoping the legacy code doesn't crash harder this time around. if 3.6 is actually faster at rebuilding ancient software at least i get time to procrastinate while the ai handles the engineering part lol

Well, still it’s crushing when I’m working on multiple projects with Gemini 3.6, Claude Opus4.8, and Codex Sol…. Somehow, it was spending more than 50GB when it has been frozen… After that, almost of chat history was gone! Did someone steal it??? Or, is it just disappeared??

Please, could you add a separator to select the thinking items separately? Otherwise, seeing so many models in one list looks very messy due to the repetition.

Use it only for designing and ui/ux fixing Works well for that.

When AI agents start managing the legacy code, the real win isn't just speed—it's how they handle the messy transitions.

Gemini is the worst for coding

I'm seeing a reproducible issue on Ubuntu 24.04 (v1.107.0). After a large Next.js analyze/redesign task, Antigravity now hits 100% CPU on every workspace. language_server loops and TS reloads continuously. Happy to share logs if helpful.

Add sonnet 5, add Opus 4.8

Damn, it only works on your machine, holy shit!

Big flop

I analyze with several prompts and gemini 3.6 flash is better and fast.

Why don’t have flash lite on antigravity?

Anyone used ? Please share your honest review here 👇

velocidad no es igual que calidad, Gemini 3.6 flash tendría sentido si tuviera el precio de 3 flash, aunque fuera 2 o 3 veces más lento, de que me sirven 350 tokens/s si la gran mayoria de cosas no lo hace bien?

I have a suggestion: Like codex, can ya all remove the 5 hr limits? If you do, i'll actually be tempted to use antigravity and make projects i never did before cause i dont wanna prompt "continue" again and again.

It’s a model that’s only good in terms of speed; other than that, it’s a disappointment. 🙁

Gemini 3.5 Pro 还能出来吗

Omg!