#vibe-coders
1 messages · Page 3 of 1
Also, I think in previous versions of Gemini cli, we were able to actually input videos as prompts. But now, you cannot input videos as prompts, which is one of the things I don't like. Gemini is a multi-model and shines best when you give it references and video reference was very good, but it does not allow us to do any kind of video reference in either Gemini cli or Anti-Gravity.
hey! i was wondering if you guys could intergrate a notes/ boxes/ arrows/ annotation method .. thankyou 🙂 awesome product btw!
Please don't ping Google employees here for your issues
This server does not provide official technical support
Hey @simple lindenplease keep in mind that we don't allow advertising as per rules.
anybody else getting this with flash-3-preview? "the model is overloaded. please try again later"
you can use NotebookLM
it will take videos
I read they integrated NotebookLM with antigravity so
that might be a workaround
I get that message regularly @split basin, especially first thing in the morning on the US East.
Yeah, I hope they fix it
Ever since Google launched Nano Banana and Gemini boomed, that's the condition. I wrote a correction routine in my LLM API function to handle the error.
Waited 3 days for the Claude quota to renew, and it renews at 40% and I can use it again in 7 days? lol
hey man, don't mind me asking but where are you tracking your usage? I.e. what app is that
extension is called "Toolkit for Antigravity"
I was there in the glorious first weeks, Gandalf
We had infinite Claude tokens
and I coded with opus 4.5 all night long
I miss those times 😭
I knew back then that it would never last. So I took that time to really learn a lot (I'm not a coder and have zero experience before december)
I really went all in lol
had 2 weeks off as well so
got the pro plan straight away
still feel it's pretty decent value
with 2TB cloud included and gemini pro on nice apps like NotebookLM
I was sleeping 4 or 5 hours a day squeezing every drop out of it, but it was worth every second
Bro I was on the same grind lol
glorious times haha
and I knew nothing
I just threw courses, links, and even videos in notebookLM
then I made it hold my hand and help me set up anything I wanted
pretty amazing times for self learners atm
I’m building a PWA.
Do you guys have any cool ideas, badges, or small UX tricks to show users that it’s browser-based & safer (no shady APK / EXE installs 😅)?
These days the internet is full of sketchy apps, so users hesitate to install anything via Chrome and prefer only Play Store apps.
But PWAs are actually much safer, I’m just not sure how to convey this clearly to users.
users aren't hesitating. They're installing mfin clawdbot that has full terminal acces, and they send messages to it via whatsapp.
the security risks are so obvious
Hello to all
I have created some creative web apps with Google Ai Studio and i have deploy them through the cloud
I want to remove my api from the code and integrate the code to authorize every new user to log with his google account and use his own api
<@&1009526435276394496>
Thanks for the ping
Always for the community 🫡
🫠
where can i see my usage in Antigravity? i have a AI pro google subs, i want to check my token usage
toolkit for antigravity extension
Anyone having Extension issues here in Antigravity?
windows has blocked it for me today. i cant even open it https://x.com/Lizardhunt_/status/2016453203064139848?s=20
i think it might be because of it not having a Digital Signature
Antigravity - i have a bunch of conversations under my workspace is there a way to delete them as i want to organise them under agent view haha
use Agent manager haha 😄
https://discuss.ai.google.dev/t/how-to-delete-conversation-antigravity/114671
Thank you sir!
Hi,
I would like to know if anyone using the BMAD method with Google Antigravity and if you have good results so far ?
antigravity still good dude?
loading so loooog
what happens
now there are 2 days limit reset for gemini models too
I'm curious, what are you using ?
hello everyone, do not pay for google antigravity pro plan, since it has weekly limit.
if you planned to use it, make sure you only used gemini models, claude is not available with 5h reset cycle in antigravity
hi every body
anybody interested for a better notebook lm
if you are, I have made it
Hello buddy, since you have build this using ai studio this would be simple. just prompt for ai capabilites ask visiter to put their api key and use that one..the rest would be handled by ai studio itslef
btw where have you deployed ur wb apps?
Hello everybody. Actually I want to contribute to open source and the thing is I only know c language and nothing other than that but I have also covered advance topics such as file handling, dynamic memory allocation and all . So what do you all think about my situation and also suggest some advices.
I was making a tool to generate color palettes with Gemini and accidentally discovered an optical illusion lol
stare at the X without moving your eyes for 2-3 sec and be attentive to the color palette's colors
if you manage to stare without moving your eyes it turns gray like this
I stared it for a whole minute, but I don't get it.
odd, me and my best friend we see it almost instantly, like 2-3 sec max
make sure to stop your eye's from moving, humans have micro stutters with eyes to prevent this exact issue of losing data
stare at the X like it ows you money
this works
This is a scam.
I've seen similar websites with different domains several times this week. Don'e believe what you see in the image.
anybody else's google ai studio not working properly ?
idk but I can tell you ever since Google put me on a 6 day strike with Claude I havn't been productive at all
I havn't insulted an AI that much since GPT-3
I am beyond pissed off
I used to say Gemini is about par with Claude, I unintentionally spread misinformation
Gemini is tremendously worst than Claude
On several levels, doesn't respect rules, terrible at navigating new codebases without context, hallucinates like crazy on long conversation, makes stupid and obvious choices
I'm baffled, I'm stunned, I'm disgusted and repulsed
I used to get mad because Claude couldn't figure out a shader issue on a custom game engine with 2 different rendering pipelines
And with gemini I get the same madness, but over the BACKGROUND COLOR OF A BUTTON
If Gemini would be an employee, his name would be Fired
Last name Immediately
Yes. I feel much the same. I find that claude is better than most other AI models. But their pricing really hurts 😢
This is the reason why this month I decided to try antigravity, and yeah the IDE is decent, but I'm going back with Claude Code and VS code because there is no way in hell I'm getting stuck with Gemini
Do you have a beefy computer?
DDR5 with "medium" spec, but yeah decent
Try out ollama 
I use it, sometimes, but not on what I do that requires more than 3 neurons
There really is no replacement for Claude other than Claude
Yes 110%
I have found that ollama is having cloud hosting. I think will i be using that in future
Because they have a big about of ai and will only charge for the cloud, it must be cheaper and give a better experience
I personally do not believe inference will be an issue in the future
For a multitude of reasons, but most of which is pending technology upgrade
Yes Tru
One rule I set for Gemini is Do not write comments.
Simple enough, one might think... Claude follows this perfectly, the code is ultra clean.
Gemini, well
The thing put god damn comments everywhere, had to ask it to remove them
I haven't actually seen this one before! It's a combination of the dot one and the other moving dot one
yeh I have it and I know
and gemini is just bad at coding atm
I still feel it's kinda worth it
pro plan
just hope they fix the coding model's behaviour
I don't understand these cooldowns. I've had three days cooldowns and now I got a cooldown of a couple of hours. I'm on a Pro plan that I got when i bought my phone. Will upgrading to ultra be a significant change? I asked gemini and it said it would be just a little better than what I've got now, which is not worth paying for. But if I could vibe away a whole day it would be worth it
yow if you wanna code, don't spend 100 on google ultra
get claude
you can still use it in antigravity with the CLI terminal
you'll still have the nice antigravity file browser & stuff
it's just the better option if you're spending 100 a month & want to code stuff now
imo google plan is tolerable because you also get the 2Tb cloud and some other things
pro deep research on the notebookLM
i really like that app
I read his comment and I was thinking exactly that
Gemini is just head bashing, you work extra hard for extra long period of times without results
Gemini implements 1 features and breaks 3 more
yeah
Last time I haven't feel that unproductive was back before I started using Claude on the daily
100%
I would keep Gemini as a side dish to use as a google replacement or make quick and easy web tools but nothing complex
it's not even close atm. I REALLY hope they fix gemini models and make them more reliable and less prone to going completely wild.
Unintentionally timed answer
Yeah Gemini needs a solid kick because right now it lacks
it just doesn't
I've been trying to constrain it with rules & .MD workflows
but nothing works tbh 😄
Claude feels like AGI yes, if Claude had eyes I believe it would be absolutely nuts
If you wanna vibe code or not-vibe code
anything code
I can't advise spending 100 a month on google plan, when there's a 100 a month claude plan
Oh no hell to the no
You get Pro plan max
Ultra could be worth 50$ I wouldn't get it
Gemini is trash - Sorry Google
I agree. Sorry Google
google has great AI apps
Sorry Google
true
frfr, Sorry Google
I'll use nano banana to generate any gfx tough ❤️
at least its img gen is good
it's really good
I have no clue
I also have photoshop
usually that runs the latest version of nano
AIght, cause I tested a couple times and it felt really trash and I thought either they don't use nana2, or even nana2 is trash
Either way, disappointed
didn't know there was a 2
Yeah it's the "good" one, the other one is ... meh
&like this?
You use photoshop?
i have full cloud yes
this is nano banana yes but I do not know if Google uses nano banana 1 or 2
because they dont say it of course, why say it when you can make people just guess
the Pro plan isn’t “still just Nano1” in theory. It includes the newer Nano Banana Pro model, but behavior can vary in practice.
I'll check in PS
that's what I thought but also when I go in the AI studio mode nana2 is locked being API pay wall
so I thought if we could use it though gemini's app we should be able to use it to an extent at minimum on the AI studio
but we can't, so now I'm conflicted
you only get very few api tokens with the normal "google AI" plans
at least you get a few
Guys. Being a noob at vibe coding. This feels like silly question but. Is having Gemini pro good to go with Antigravity?
it's a good start
I have gemini pro so, will I have any limits?
I got none actually, I couldn't even test nana2
but the gemini modelms are atm terrible to code with
so claude is better investment
same lol
Oh. Thats bad
google does give you quite a lot for 20 euro per month
but for coding, it's just bad atm
Yes, but I recommend using Claude
I got Gemini for free 
im gonna switch too next month
not anymore
thats the lower version
you still get 2 tb cloud space and pro in notebookLM, you get google stitch
If I would have purchased a premium then 110% I would have gone with Claude
Nah. The Gemini 3 pro
if I want cloud storage ill pay cloud storage
you can't sell me a car by offering a free cap
if you serious on coding
if you wanna make something that generates some cash
get the 100 euro claude max plan
yes
it's probably the best value atm
100%
110%
1/0 %
division by zero
is infinity
Nah. Not Defined
yeh it's undefined
was it not infinity
you gotta take the limit to zero from a certain direction 😄
only if you go to zero from positive
so if u negative dir, you will get - inf
lol
Go get Claude
yeh google calc is just "simulating" the result you want
ffs
Technically it's -Infinite
you can still use antigravity with claude
Technically it impossible as well 😌
personally as soon as my pro plan expires for google im going bacc to claude, so, tomorrow
claude + vscode with the claude extension
Oh. Dont leave GDC though
was basically antigravity without having to install yet another fork of vscode
Continue is also a good extension
GDC?
I use it with ollama cloud
Google developer community = GDC
ah yeah no, in case Google team decides to do a meeting and actually fixes things I'll stay here
Lol 😆
what are the chances that will happen?
1/0%
lol
but I have low doubts in companies that are too big, its like buildings, earthquakes at a higher frequency for smaller buildings, more happens in the same time frame when compared to big buildings
Google is just too giant to be able to tackle AI for coding at the speed anthropic goes
which is why they attempt to diversify their product
they shove gemini everywhere
Yea. Thats very True. And they are focusing on implementing AI on all their product specifically
yes
Not for coding. So there's very less chance that they focus on coding
yeahs its in mails too now
What do you build btw?
because I'm limited to gemini now my personal debt manager (offline, simple) and it can't even do that
Lol 😆 🤣 😂
with claude I was making my own game & game engine + my own programming language, then I made claude convert my game & game engine to using my own programming language
Gemini could never
the thing would crash thinking about it
Really???
yeah I know it sounds like fiction but yeah
Then what you did HR?
Yes
not sure I follow where human resources should fit in all of this as a question
I just have a workflow setup that makes handoff tokens in .json format
Can I try your programming language?
no, it's still WIP, because of Claude I'm able to quickly make changes, but soon you will yes
what will be the name?
nano, nano lang with .nano extension
WIP == Worst Possible Person???
work in progress
yea
btw, some token saving tips for antigravity users:
Don't use "prompts". Make "spec sheets". You can use Clade chat bot to make a complete, detailed spec sheet document for free. THen you can input that in antigravity claude opus 4.5
Tanks
what about "Tanks"?
Thanks = Tanks.
yeah they do
my lang is based on rust, it transpiles to rust, so you can take any project, run nano-convert and it renames all your files to .nano and modifies the project to add build logic for .nano files
and from there you can either code in rust, or use the added syntax embedded within nano
lol
its like React but native to the language rather than a layer on top
you can go to https://claude.ai/ and talk to the claude chat bot, to flesh out your prompt
among other new exclusive features
use claude to prompt claude
Dont you code yourself?
what i do is i five promts to chatgpt to modify it and then feed it claude
or better, just make a specsheet.md file you can reuse later
Only Vibe coding?
eeeuw touching code yourself?
atleast i do
that's not really "vibe"
Oh my goodness
haha
I only input my API codes & other secrets in the env files
i do both
If an architect design a monument there's no reason for him to build the whole thing
that's how much I touch code
maybe bash some terminal commands
I know nothing about code for web or apps
only for PLCs and robots
I touch lots of code just the trash one.
I design, claude does, I supervise because models are trained to take shortcuts to "complete your request" even if it means cutting corners
That I build
the models are plenty good at coding, you just have to do the engineering
Um. Thats a good way to go. But what's the vibe in not coding yourself
my father used to code with machine language. He was reading instructions and adresses in hex code. Line after line. That was before higher code & compiling was invented
After all we are software engineers. We love to use 10hr of time automating behind the task which takes on 1 hour
like, no one does that anymore now
the same will happen now with AI agents
no one will touch the code anymore
once you've made many parsers / renderers / things like this you realise the only difference is the execution/architecture and the rest is all boilerplate
we will just level up
Yea. Thats a real advantage
No boilerplates
and the architecture is what is hard to design for AIs (and humans tbh) so you have to first yourself make a concrete plan, once concrete is set everything is easy
that's what I'm trying to make
the system to make things with
not the things
its like making a game with an already sketched out idea with everything planned
or attempting to make it along the way
making it along the way would be a mega mess, uncoordinated bugcluster
unless you make like flappy bird ofc
Dear Diary,
From now I will vibe code (my way)
you saw the geoffrey huntley video ❤️
possibly
He's a prophet
but it's hard to use in antigravity
because antigravity has even stricter compaction limits
it starts compacting after 7000 tokens
so you can't really do the ralph thing in antigravity
without api and containers :p
for ppl with money, I already have it set up
thought about setting up GLM offline to attempt a ralph loop
but I'd need claude to make the framework quickly
and anyway all of this is low in my priority list
I'm trying it with gemini flash
the context is inside files and folders
yeouch
flash is much dumber than GLM you should try with GLM if you have a decent GPU
https://www.youtube.com/watch?v=16XsmVLE0-c
I already have the loop working
Hey guys. Someone on reddit asked how my setup worked, and at the same time gave me an insight on how I could make this ralph loop thing parallel. So here it is. I'm basically attaching 2 antigravity instances to their own containers. Have them make a chatbox together with one doing front end and one back end. Small test for now.
in parallel
but it doesn't clear the context after iterations
nice to see, though the example is small and not complete
would have been nice to see the results more, but good enough
it can't be large, because i haven't fixed the context thing
with claude code
it just starts another instance
I can't automate that with antigravity
i mean in window size, had to full screen the video to see results
It's because I run uwhd screen
I recorded OBS on bit lower res
I see
it generates videos of insane size otherwise 😄
yee i get it I used to have some
i bought the first gen of 32:9 when they came out (LG)
I like playing euro truck simulator
and use a webcam for head tracking
it's so nice
returned it after a week, cost me 144$ to send back lmao, but I was disappointed
I've had so much fun & immersion with it
but I do play with steering wheel, pedals & webcam headtracking
so the camera follows my face
went back to 21:9, then over the years moved back to 16:9
now I have 16:9 in a quad arrangement
it's so good man
I love it
also like farming simulator
I use steering wheel & joystick with that game
I love many games
Bye guys. See ya tomorrow
have a nice evening
I just don't like game engines, they aren't made to handle 32:9, it's the only reason I went back
Making my own game engine now I need to buy a VR headset again (sold the one I had) I have many things to try
nowadays they do
yeah modern ones maybe yes but most games are already releasedf
yeeeaaaahhhhh but still
if I had to go back id go 21:9 but defo not 32:9
i thought about it and my next setup would probably be 3x 21:9
im only now looking into OLEDs because I am exhausted about pixel ghosting
now that OLEDs aren't entirely trash anymore
thinking about 1440p (not too low, not too high so GPU can handle) and probably 250-450 hz, not too low, not too high (for the games I play at least, most people would want more like 4k at 120hz)
though my game engine manage to pull 1700 fps at 2 million triangles in full PBR I'd be curious to see that on a very high refresh rate OLED
just to know what it feels like
IPS is ok?
doesn't have burn
nope, none is
burn ins are not an issue anymore, its pixel ghosting
or just ghosting people call it, it's been a curse of mine since 2019 (when I started noticing it)
meh my monitor is perfect as far as I can see
X)
I built this PC during covid
was so hard to get a gfx card then
it runs RDR2 on ultra with all options
my pal did the same, told him to wait, he couldn't
I leterally couldn't
he paid sooooo much for a GPU lol he regrets like crazy
rip
dude what the hell, gemini attempts to check another project without being prompted
back in 2020 that was pretty baller
click the tripple buttons next to the agents, go to settings
tell it it can't touch files outside of workspace
I opened a file of the other project into it just to examin it and it decided to just freestyle and read the other project
bro just
don't run it local
get docker desktop
run it attached to a docker container
its fine I just pressed Deny, its caught
also its just for 24 hour after that im back on Claude
Gemini can hell inrot
it will destroy you if you use gemini 😄
don't use it local
run a container
be safe
i hate containers
I ❤️ containers
i only use them when required, otherwise I run without, I don't publish to repos anymore, I don't even use github anymore I use git locally, im so done with all the tools slowing us down
people dont realise but they spend 50% of their time fighting their tools
I only git projects
atm I'm mostly working on my dev setup
"oh because our app is contained and this thing is locked up behind this and this blabla" they end up spending 3 days making something I can just copy paste or something its crazy
I usually run agant inside and outside container when I'm setting things up
I've shown you what kind of degen things I'm trying to run
that needs to be in containers
the containers are running locally on docker desktop ofc
i just run locally and hope for the best, even if it would be to delete my whole project everything I've learned along the way I can just make a better one from scratch
from the lang to the engine and everything in between
but I trust claude not to mess up anyway, gemini, meh, but its for a day
bro, don't trust gemini 😄
holy jesus christ you were right
what a coincidence
gemini fkin went OUTSIDE the project and started modifying another project while I was playing gunz
what thef
note : DO NOT open another project's file into antigravity even if its just to look at it and its not in the project my god
gemini is the worst, THE WORST
lucky there's an undo button
bro I told you
don't let it touch your stuff lol
it ignored my gitignore
this MFer
it's a wild horse atm
I've been trying to tame it
but you can only give it simple task
INSIDE CONTAINER
so it doesn't touch your other stuff
It's very good at generating YT thumbs
letsgoo
I think I'll just finish the touches I have to do on my app right now and stop programming entirely until claude
this was the last strike for me
gemini is dead to me
yeh sure
neat
the third one and last one are the best from the bunch u sent
though last one feels AI generated a lot
they're all very much nano banana
yee i get that
did you gen this using "generate someone who thinks is cool"
don't tell me its you
I copied the "ibiza final boss"
#electronicmusic #dancemusic #festival #dj #producer #shorts #ibiza #comedy
this guy actually exists
I actually look like this
never watched startrek & star wars, always telling myself "ill watch em soon"
now its "im too late " lmao
lol I better remove this
yeah lol
haha
Thank you for the ping!
What is your programming level if we delete the AI layer
Just so I can propose something in the lines of
Is Level 1 or Level 33 higher?
Hi @faint aspen
Please avoid sharing posts or your own links in chats as it qualifies as self promotion/advertising - this is against the server #rules
Hey am I allowed to share a link to my game here?
Doubt
Especially with what the mod literally just wrote
I found the best way to make an app, I'm slashing through days of work instantly.
I made a README.md with precising styling for the button, the scrollbars, and everything else UI/UX I desire with precision using another app I made as example.
Then in DO.md I just list what I have TODO/TOFIX/etc and I tell the AI to read DO.md and that's it.
Basically a ralph loop - but without anything extra
Crazy, flash is actually decent now
Gemini quote of the day Apparently, saving to localStorage solely isn't cutting it, which is painfully obvious now.
If you want a proof for how stupid Gemini is, here's the latest flagship
"Set the title of the app in full caps"
- Decides to change priority to NANOPRIORITY
my god
can't even do that right
This channel has gotten very silent, really shows how most of the traffic on antigravity was because of Claude, and now that this is gone because of hard limitations, the traffic is also gone
Or... most people, like me, are just busy building stuff with AG and don't have time to chat.
so hey! gamblerss!!
Saying you don't have time while an AI does the work for you is the move vibe coder thing you could have said
I focus on what the agent is building.
yeah that's why we're here
Thats right, Gemini on antigravity feels very bad, and the límits even with pro plan are imposible to real work, only for hobby use! Hope this change soon or no one use antigravity in a couple of months
Yeah already cancelled, plus I just tested OpenCode and not only is it free, it's as good as Gemini
Gemini, to my eyes, is dead
I'm having so many issues with Gemini that even though I paid for it I'm using OpenCode right now and I'm having less issues than with Gemini. Crazy to say out loud
I'll go back to Claude probably though
Looking for resources on building voice-enabled AI agents.
I have a chat-based assistant in my app that helps users with questions, story craft, and navigating platform features. Now I want to take it further and add voice interaction—speech-to-text input and text-to-speech output—so users can operate hands-free.
If you've come across any tutorials, articles, or videos on building this kind of interactive voice agent (especially within a web app), I'd love any links you're willing to share.
I used to use WhisperX I think it was called, its like whisper from openAI but it's faster and gives you timestamp for each syllable/word if I recall
For voice to text
Text to voice I always used cheap models or cheap solutions because the only good option I had was much too computationally expensive (sesame)
I was able to get Cursor to help create interactive features and functionality that activated via voice command, but it required me to map out all sections of my platform for routes. But then that made me wonder why my chat agent can help walk people through the platform via chat, but I can't get an interactive agent to see the same routing.
I was actually trying to use page readers to help with some of this to make the site accessible.
Hello, popping in here again as I got good feedback last time. AILANG is a new programming language aimed specifically at AI coding that can be used from Antigravity, Claude Code, OpenCode etc. it recently demonstrated 10% fewer tokens and a +6% task completion vs python in its own benchmarks https://ailang.sunholo.com/docs/benchmarks/performance - the aim is AI using this language creates more stable, secure applications with less tokens and cost.
With AI doing the coding, the human interface is moved to an Agent messaging system (with imports via GitHub issues), design docs setting up data contracts and monitoring of activity via an embedded dashboard https://ailang.sunholo.com/docs/guides/collaboration-hub
I should get concerned when even the documentation looks AI generated
Is this AI delusion to the next level
Oh my god it is AI generated there's Em Dashes everywhere
Dead internet theory here I come
sooner or later, everything will be automated by AIs and human just sit and watching 💀
Dude, it's already like this
You know what I do right now
I run this on a loop:
Read README.md, then complete one task from TODO.md.
Once the task is complete, please notify me and specify the completed TODO item. This detail is required to begin functional verification.
Once I confirm the issue is resolved, remove the corresponding task from TODO.md and commit all current changes made by you or someone else.
I basically do nothing, but test, say if it works or not, everything else is automated
Yesterday the AI prompted ME for a question I was like
It picked a TODO task and prompted me how I desired to have the comments and it gave me multiple choices... The turned already have tables
does anyone know how to add a API key to antigravity?
it uses GEMINI_API_KEY, I think.
Oh, apparently not. it just uses your current user credential
why doesn't gemini api support external tools along with built in tools like search?
It does.
You can request Google Grounding, and provide a list of tools.
that's from the API workbench
But you can find the same interface in the low-level API calls.
@cosmic ferry ^^
so i can use grounding with search + any tool i create myself for my app?
As long as the model permits it
That was gemini 3 flash
but I think most of the Gemini 3 can do google grounding now
You can play around in AI Studio to get sample code for your language.
I didn't know Google hosted a clone of Claude for inference for antigravity
I was wondering why Google even offered Claude at all but now it makes sense through their service Vortex AI or whatever their codename is
Hey all , I have a doubt does anyone know how to take free openai api key for the development ? is there any platform providing that ?
Well I do know Google offers free APIs if I recall, I used to make apps with it back when I was poor, even audio calls which is mind blowing to me
It used to use flash 2.5 if I recall or something like this
As far as I know the others aren't so kind
look at this some guy in china has 6500 accounts on antigravity , how the f is it even possible to create that many accounts
lmao😂
how do u know its real?
damn, ppl believe anything they see on internet today
I have seen enough Chinese content to not doubt this, really
6500 accounts is nothing, do you know how many people there is in the world, at least 5 people
I mean, sure even if its true, maybe they have found a way to automate the creation of several accounts perhaps, they are exploiting the platform for some use
Google accounts are used for many things, there has been a hell lot of google accounts floating for long enough for Google's system to not know which is real and which isn't
How do you think some shady services in regards to view count and everything similar on youtube happened
Humans
The only thing that would fix these issues is an internet ID per user required for registering, but this is like the debate of the last 30 years, this nukes privacy at the cost of security
That's what north korea does, so they have no privacy but at least there's barely any cheaters in their games & services
Because if you get banned over there, that's an internet life sentence lol
Holy smokes. My only regret is not upgrading to Ultra earlier
🎶 Its a new day, its s new life 🎶
😂 lol
168$ a month, per month for 3 months, and then 400$+ a month, per month, no thank you
Claude Max 20x for half the price
You see the list of Google people on the right, those are probably paid by the extra 200$
I'd rather get Max 20x and give 200$ to charity each month
Does anyone here use the Gemini API for creating Nano Banana Pro images? (the gemini-3-pro-image-preview model)
i once use the gemini api for creating images using gemini 3 pro image
Is this a good channel for antigravity issues? The latest update prevents attaching running containers on WSL and MacOS. My ArchLinux environment does not have the issue though.
This is indeed the right place to discuss Antigravity
I've reached out to the reddit community and they seem to facing the same issue. I wonder if anyone here has too or if there is a place to submit an issue.
Based on the logs, these are the steps that succeed and then fail:
┌─────────────────────────────────────────┐
│ Antigravity Server Installation │
│ Location: /root/.antigravity-server/... │
│ Status: ✓ Already installed & running │
│ Listening on: 34023 │
│ Token: ####### │
└──────┬──────────────────────────────────┘
│
│ 4. Port Forwarding Setup
▼
┌─────────────────────────────────────────┐
│ Forwarder Process │
│ • Create forwarder port (62135, etc.) │
│ • Attempt to forward 34023 → local │
└──────┬──────────────────────────────────┘
│
│ 5. Execute forwarding subprocess
▼
┌─────────────────────────────────────────┐
│ ❌ ERROR POINT ❌ │
│ │
│ "subprocess terminated immediately │
│ with return code 127" │
│ │
│ → Command not found │
└─────────────────────────────────────────┘
I don't think the log says what command was not found though.
Just released Opus 4.6 looks incredible! I'm not at my PC. Is it available in Antigravity yet?
Not yet:
I am new here, so this is a question for everybody, is there a way to use third party models like GLM 4.7, Kimi or others inside of the default Antigravity chat? The goal is have higher quotas with a cheap reasoning model, but also keep the browser capabilities to click, analyze and reporting enabled by Flash.
Was this discussed here before or attempted by anyone? Open to work on this mod.
hi
It is in co-pilot.
Hi, I have a question. Is it normal that once I reach the token limit for Claude Opus, it makes me wait until December of this year? 😭
Date in US format could be Feb 12.
02/12/2026 in the US is the same as 12/02/2026 in most other locations.
Man AG sucks now, im moving back to codex
antropic just release opus 4.6 and also openai just released 5.3 i am thinking after this what google is cooking and will release in new few weeks 🔥
Pinning this again - Anybody have an idea?
I do hope google add support for BYOK, cuz i also want to try kimi in AG 🙂
Its the orchestration that AG does very good, otherwise the default google models are really not good and I've had multiple instances of Gemini deleting entire working parts of the codebase
Did you see BYOK mentioned elsewhere I can participate?
Actually, you can use Kiro, but it's a bit out of the question; you can't use Opus, but it can do everything automatically.
I made a Discord Bot using GEMIN 3 PRO. Here's what it can do:
Slash Commands
- Brain Module: /ask-silk, /idea, /roast, /whois, /summary, /define, /slang, /translate, /ship
- Chat & Voice: /chat_toggle, /voice_mode, /persona
- Creative: /imagine, /tech_news, /voice
- YouTube (Shifo): /stats, /latest, /shoutout
- Architect: /architect, /demolish
- Moderation: /kick, /ban, /unban, /purge, /slowmode
- Utilities: /ping, /uptime, /serverinfo, /userinfo, /avatar, /calc, /poll, /qr, /dm
- RNG: /roll, /flip, /choose
- Fun Text: /mock, /reverse, /clap, /say
- Help System: /help, /quick-ai, /yt-explain, /vision-help, /utility-help, /text-help, /staff-mod-help, /server-build-help, /ai-chat-help, /shifo-info
Passive Features - Global Triggers: Responds to @Mentions and Replies server-wide
- Hybrid Voice Engine: Auto-generates MP3 audio attachments (with text sanitization)
- Smart Scrollback: Reads previous 20 messages for context
- Rotating Presence: Cycles "Assassin" statuses (Watching/Playing/Listening)
- Dynamic Tracking: Live "Watching [X] Members" status
- Creator Protocol: Hardcoded ID verification for override access
- Render Keep-Alive: Background Flask server for uptime
Can we get Opus 4.6 for antigravity?
Gemini is the only model I have to tell Please stop thinking out loud I don't want to hear everything you think
Even Kimi K2.5 doesn't have that issue
There is goes again
It only got released a few days ago, wait for some more time they may add it 👍
Codex is running a 2x monthly promo now, but usually the Plus is similarly limited weekly for the GPT-5.2, just like Pro sub on Google is for Claude
at least that was my experience testing it a few weeks back
I’ve been back on codex for the past week and I haven’t even hit any limit, unlike AG now where I can easily hit a limit in like 1 hour
because they're running a promo, and depends on your model
AG back then was so good tho which sucks
AG is very slow now plus limiting. Agreed
AG now faster for me. Was crawling since yesterday. Could've been the changes with Claude 4.6 overall, hopefully speed stays
opus 4.6 is available for ultra users
Guys I made a discord bot....
It's completely made by Gemini 3 pro in Jules
So I'm trying to get my agents within Antigravity to follow workflows more reliability. The idea is that I can include the /workflow in my prompt, and then hopefully my agents will follow that flow for the whole chat, because they are creating MD working documents they check against to confirm the job is done or not.
If all I have to do is reping /workflow every prompt, that's not too bad, however my agents seem to require me to ping the flow AND give explicit directions to follow it, every single prompt.
Any tips?
ai studio is so bad
did they use ai to code the github commit part
theres no way to relink github
its trying to add the gemini chatbot every 2 seconds
theres literally no controls for github
like actually how do they not have single control for github how is that possible
@elfin viper I understand it can be frustrating but please do not evade the server filters
Would you be able to assist me with Antigravity /workflows or point me to someone who can?
Sorry I personally can't offer you that help but someone else from the community might be able to
You may leave your query here if it is something specific
Which LLM Model are you using?
is anyone else having this issue in antigravity where they are logged out each time the app restarts?
I have been having the issue of being forced into relogging into the github portion on launch at least ~1-3 times each day this previous week Feb 2-6
Mostly Gemini 3 flash and pro. I've used opus 4.5 as well, however it straight up told me "I thought I new better than my instructions that said this protocol is ALWAYS ON and I decided to just do the task without following the /workflow you specifically included in the prompt" and so I went back to Gemini and the /workflow was honored 😁
Opus 4.6, when I can't Opus 4.5, when I can't Gemini Pro, when I can't Kimi K2.5, when I can't Gemini Flash, when I can't I go play video games until something refresh
Opus is much much smarter than Gemini on several levels
And Gemini Pro is smarter than Kimi and Flash
I could run Opus 4.6 on a live server, but Gemini Pro, hell no
Claude
The only reason to not use Claude is because of financial reasons, other than this it's 100% Claude the winner by a big margin
As it is more pricy
Is claude models not working for anyone else in AG?
They severely limited the Claude models I couldn't even tell you anymore
So I'm left using Gemini trash that needs 30x the amount of prompting for the same result
I'm so tired of Gemini
6 more days and I get Claude Max, and I'm finally free from the curse of Gemini
The Claude models are working fine for me in AG. Used them for two days already and haven't hit the weekly quota. Last week, I busted the quota on the first day, so it seems they increased the limit. However, I am running Opus 4.5 high and it seems too fast, compared to using 4.6 in Claude Code, so maybe they've reduced the thinking effort. All in all, I'm happy.
Yeah they started working for me again.
for some reason i feel like gemini models perform way better inside amp code and droid cli
Placebo
i have seen some benchmarks too
Interesting
Is there a way to view how much of the context window is filled up when using Antigravity?
Also, does it get shrinked/compressed at some point, or no?
Guys, Appscript its something else. I'd love it if there was a channel for that
perhaps #workspace
hi guys! , antigravity reducing de request for google ai pro , now i have to wait 4 days to continue using gemini high and low , that's is crazy.
i am seeing the request reduced more than 50% in the las 3 weeks
how is that possible, honestly at this price, it doesnt worth the montlhy subcription when you have the same for free in firebase studio cloud.
did they fix the antigravity opus model limits fix?
or is it still 1 week now?
for pro users?
😭
antigravity down for anyone else?
Yeah
They're limiting it from now on from what I can tell
Yes, I was just about to ask the exact question....
Not sure what is going on, but, found this in the Dev community;
Not really sure what it means or what is going on, I honestly didn't know that AntiGravity was anything more than a standalone IDE... so, this is a bit concerning...
Anyone else know what is going on?
Running Antigravity in WSL2 using the Remote-WSL extension, which works fine, as this problem arises even on local Windows. Everything in the IDE is working for the most part, with wsl2 set to nat, instead of mirrored, which is the default. The problem is that the agent sidebar is stuck permanently loading, eventually giving a “Antigravity ser...
Working for me
Also as a backup option for Antigravity - I'd check with the Google AI Studio and... It's a different provider, however I find Mammouth.ai to be a fantastic tool. It asks wonderful questions for spec driven development and alignment.
it's back now apparently
Antigravity does have an independent "agent manager" that runs in parallel with vscode and the individual tasks.
You can launch parallel task execution, as long as they don't overlap.
I've found them better and more aggressive on planning on antigravity.
Anyone here using OpenClaw?
its 1 week
you can buy the ultra plan which later going to be rate limited to 1 week too if you're using it way too much
i think it's 5 hours first then 3 days then 7 days
im currently on 5 hours
Is that from your own experiance or satire? 😄
guys why do i have to wait around 5 days for token reset on claude opus 4.5?
I havent seen that, but seems bad, lets hope its bug
why is gemini so slow and buggy in antigravity when it is even their own model? i have not encountered major issues with claude models
2026-02-10 10:21:30.709 [info] E0210 10:21:30.709078 18232 log.go:380] agent executor error: model unreachable: UNAVAILABLE (code 503): No capacity available for model claude-opus-4-6-thinking on the server: UNAVAILABLE (code 503): No capacity available for model claude-opus-4-6-thinking on the server: model api cannot be reached
Getting this tho alot
Seems servers are flooded ^_^
Yeah whats with usage it doesnt reset now after 5hours also
Its kind of very bad when everything changes without any announcements and heads up. 
Yeah that doesn't feel great, tho Google still gives more compute per dollar than anyone else, far as I can tell....
i got capped for 5 days
yes fully agree, feeling like scammed after paying 22$ they changed it, without any previous msm
honestly i am removing the subscription probably, antigravity vscode is always with issues, most of the request are falling with errors due to agent, and claude and opus are almost the same tier as free users. then later you can use gemini 3 and flash for free on the firebase studio cloud....
so can i know what are we paying for? subscription model is not fixed with a proper model
i got 20 msm in a row about opus agent down, then later they told me that i have exceded the quota limit, and now waiting for more than 4 days to reset
just after 2 hours of work
now i have to go to firebase to use gemini 3.0 for free, so quite unrealistic, that free tier over there has more capabilities that current one under subscription
Free more than paid? I haven't seen that
Compared to Cursor and Claude, I believe we still get WAY more total compute for 20 bucks a month
Even though they are being funny with the limits on Claude
I saw it ask me for 5 days and then after I saw it refresh after 5 hours
Hello all
I was just reading about how there are fake skills that open claw can use that steal all of your data or scan through your computer, so watch or try to restrict what it can use... sounds interesting though
that never happen to me
yes absolutlely, go to google firebase studio, it's the same vscode but cloud, there you can use flash and gemini 3.0 for free, with endless quota and without limits. i have been using it for one year , they also send me some links to open google dev program and participe in impovements
Endless quota for free without limits? That's not what I saw in their website lol, that's not economically viable either, would be abused like crazy
well, that's how that works on my end for the last 10 months approx
i guess if you accept the google dev program terms, you are submiting all your usage and code for their improvements, so probably thats's how you are paying the service
i think most of people dont like to work on cloud , but yeah sometimes the service is down per hours
So AG keeps forgetting where i.e. my database is, and I remind it "docker.." and it goes "aaah!"
I've tried making it have a note to self, but it forget checking its own note.
Whats the best way for it to remember things
and yeah , you have the cloud limitations, docker, domains, databases, and few things are quite tricky to work with
Hello, sorry I know this is not the Channel for Antigravity, but maybe any user can help me here who has Google AI Ultra.
I want to upgrade from Google AI Pro to Ultra, are the limits higher in Antigravity?
This is the antigravity channel 🙂
Oh okay! even better :)
Do you have a answer for me maybe, it would be very helpful because the Antigravity site states just *higher
for both combined
The limits are higher, but I dont know how much. After I upgraded a week ago I have yet to hit a limit
Thank you! This is very helpful. Have you been on the AI Pro before or Free?
If you were on AI Pro did you hit the limits often so I can compare?
i love vibe coding
I was using AI Pro before. Started out well but in the end I had severe problems hitting limits on all models several times each day
Thank you I upgraded to Ultra and it feels faster. I hope I do not hit rate limit instantly
Google Antigravity - The beatings will continue until morale improves
You can try switch between fast and planning mode
Only use planning when you need to create a plan
Yes! I am not hitting the limit for now. I hope this stays the way. Switched from 3 Google AI Pro to One Google AI Ultra
But mostly I like to look into the code and make the plan myself so I mostly just use Fast Mode
You should try claude code
Buy Claude Max 20x
200$/month
Would you say it is more worth than Google AI Ultra?
I like the Opus models most, but I also like to have some gemini 3.0
Ye
Gemini 3 pro is good also
But I appreciate opus than gemini
Yes I feel the same way. Thanks for your tips! :)
XD
That's crazy, AI Pro is not what it used to be
3.0 flash?
fully agree, they reduced quota limit at least 50%
yes
use incognito maybe?
i am logged into my pro account
ok tbh idk how that works i just use the free version
and ive never hit the limit before
idk lol ¯_(ツ)_/¯
-# (on https://gemini.google.com)
-# (on aistudio ive hit it several times tho...)
yes thats the funny thing, you can do almost unlimited in teh free version , and fully unlimited in firebase
so why buy the pro version?
wait wdym fully unlimited in firebase?
exactly , not sure if they are doing that math, but pro version doesnt make so much sense now
no quota limit over there
under google dev plan
soo... cant u dos it then?
i think everyone is getting taht mail
wdym
No you cannot, rate limits apply. Just no hard limits
oh so there's rate limits but no storage limits?
also whats the google dev plan? (is it free?)
oh wait.....
quick search reveals that firebase is owned by the same company that owns the search engine i used to find the info
-# ||(ok so yeah i didnt know firebase is owned by google)||
lol
yeah , firebase studio is google vsc in the cloud
ok i think ur on mobile
vsc? like vs code? isnt that owned by ms?
yea but like
doesn't microsoft have copyright over it or smth
no idea sir, i am a simple man
-# same here xd
also what does Bronze (Lvl 1) mean here
finally people are getting parma banned for abusing
Why earn Lua? Come on a journey through Lua's essence, creation, strengths, and use cases.
Lua stands out for its simplicity, portability, and versatility. Designed as an "extension" language rather than a standalone option, its minimal footprint and seamless integration with C++ have made it the go-to choice for developers looking to bridge us...
I think this sounds very useful, almost like I have very little utility for something like python. I would recommend Lua
I think GCP is in a transitional state of development, it might be a bit slower to configure, but I imagine people using Lua or other similar languages, like Facebook probably already used Ruby I imagine.
Google mightve used Go, im not sure, to provide AI in some ways I imagine, front end. Ruby on Rails does that
Facebook doesn't market AI really
they just need to send back the quota to the pro users now 😄
im kinda fked for the next 3 days, wondering about ultra plans, but i am not in situation to spend 240 monthly
do we have on here any google team mate?? maybe they can help us a bit with some discount for 3 months >)
Before I upgraded to Ultra, I had good limits using 2 Google AI Pro. Its worth more than Ultra if you dont wanna spend 240
good idea, noted for testing
thx
I had a hunch that Ultra was terrible, which is why I never got it
plus their price increase to insanity after 3 months which is just, yuck
Hello,how many of u like gemini cli
it kind of always stuck in loop
claude opus 4.5 is far ahead of generations,i dont know why
yep
Yes I would argue that it is not worth the price after 3 months. You get more using 2-3 pro Accounts for much less price
I mean it's both
No one wants to get used to their tools just for them to skyrocket in price
Especially considering there's an even better model next door, which doesn't change in price
which one are u talking about sir?
Claude that is
I think leverage does that, I have a good reason to explain why in terms of accounting. I mine gold, due to my physical labor unless I were to rent a large machine (probably a large liability) i can only acquire so much volume in the specific operation.
I think a lot of companies struggle to see that with our equity system and the dialogue
Its also external, with most small business being more recessionary. Which is what makes me wonder why they even expect that initially
Its actually regulated by law btw, which i always felt was weird. Like you sell something..
This is different though, the issue is not maximum volume capability, as the usage doesn't change, only the pricing does after 3 months
I do understand the maximum computing capacity that companies have to deal with, which is why Anthropic is more restrictive in terms of tokens as well, they have much less computation available than Google
And Google does have advantages to offer, like stability, their models is much more often available than others who keep crashing during tweaks and updates, and also usage is much greater, but at the cost of using a model that still eats crayons if it likes the taste
I mean I was talking more about the expectation, but i think the issue is with subscriptions. As an individual, it makes me afraid, they say whatever, people just walk all over me
At least I feel they were doing that, maybe I argued with them, like I even know that.
Afraid of what?
Subscriptions
Well you can just subscript and unsubscribe instantly, it'll only charge you 1 month
I know Google does that..
They all legally have to do that
All subscriptions are legally required to be cancellable instantly
I dont see that
Without losing the benefitrs of what they pay for
Its possible indexing is hidden
It is possible yes but there's probably laws for that, but I tried with and without, so far both has no difference
Idk why, just didnt see it all the time. Variable, I mean it was more difficult in between things
I mean..
Theyre all gona start marketing with it, I feel..
Or whatever..
With what
With subscriptions?
AI is a service, so of course it's a subscription.
They won't sell the cows if they can milk it.
If you feel uncomfortable with paying for a temporary product you can always use free models, depending on what you do some of them are very capable, some can even run on your GPU easily
like qwen3:4b-instruct-2507-q4_K_M which is only 4GB yet is capable of agentic tasks
Thats a good point technically, but imagine if you have actual robots like a car. Well you're probably not gonna charge a subscription on the AI necessarily.
And GCP sell the data, as far as I'm aware, but thats like the tertiary market or something of it.
ChatGPT I think is saying theyre not making money and its creating debt, which is bad for everyone I think.
Oh they definitely would sell you subscription for those
Remember we are in a capitalistic world, everything revolves around money
If they can make more they will
You sell the robot though, and theres multiple other requirements, like energy, repairs, etc.
Not only that, theres just not pure AI in it
This is because other companies know how to make the robots
Other companies do not know how to make a Claude Opus 4.6
Or else they would be as smart as it
But again, free models are catching up quickly, Kimi K2.5 is actually decent, and comparable to Gemini Flash
In a year we might have models as smart as opus 4.6 for free
But Opus 6 will probably be even more insane
I mean theres incentive for a robotics company to allow it to work with computers
But again, you're comparing two entirely different products
AI and Robotics go along together, but they're not the same product model, and in a capitalistic world they never offer the best solution for the customer but rather the best profit solution for themselves
And this applies to all companies, no exceptions.
I dont think thats capitalism, I think its consumerism, recently all the pawn shops changed and the assets have no where to be sold. I think that opens a huge incentive for fintech
Consumerism is okay, its adapting, ive always been more about like variability and higher margins. All kids are, idk its better.
I see my own bank account, I have my own gauges, I dont see others
If the pawns shops didnt like fully change their business model and create market turbulence, id just agree more outright
It's not consumerism, as if we had the choice most people would want to own their AIs
It's capitalism
Consumerism drives capitalism to an extent but that's a different topic
Its like food and commodities, but gold has different functionality as a commodity and uses bonds.
Hm. But I'm not interested in talking abvout consumerism anyway, I hate humanity and this is one big part of the why
Im sorry to hear that
I think the point is to analyze it, to try and help your own infrastructure. Sort of inherent to the measurement itself
Semantics maybe, and tbh I have never heard anyone use that in a positive way, idk why its not.
I dont remember the definition exactly, or the concept atm
😁
yeah usally happening to me as well, i feeling tired just seeing so many txt, would be great to add a summarize windows in discord, so we get like just 50 lines to reaaad per day.
probablyy it s a good idea for an extension project
Let me planning for a better discord plugin =))
Hi , everyone
Anyone here from India who is good at vibe coding?
I'm good at the latter
I think politics is not allowed in the server but I just wanted to say your status is based
Thanks
Once you tell Gemini to stop thinking out loud it's decent to be honest
Also I noticed a hack that works on Claude but not on Gemini for some reason
In fact Gemini performed worst than without that hack
I precised "Do a proper implementation rather than a hack around." which on Claude works wonders, the model takes the task even more seriously and takes "the hard but faithful path" where Gemini instead did ... terrible, basically the invert of what Claude did
It picked a wrong goal instead
eyy sup guys!!! i got back my quota limits and im happy >)
what are you coding today?
how do i fix this?
whenever it runs terminal command it will just show this "running" forever
looks like overloaded , try to restart, i never got that issure at my end, also try to update antigravity, last update is with opus4.6
i think i found the issue:
https://discuss.ai.google.dev/t/antigravity-agent-terminal-commands-fail-to-execute-due-to-missing-enter-character/121042
Report 1. Description Agents operating on the Antigravity platform are successfully sending text strings to the background terminal, but the commands are not executing. The text appears in the terminal input line but remains idle. This issue appears to be caused by the models failing to append a newline/carriage return (↵) at the end of the c...
can you imagine
fix this asap
or actually downgrad instead of upgrade 😄
https://discuss.ai.google.dev/t/agent-terminal-deadlock-stuck-on-running-for-all-commands/120417/10
downgrading to the last 1.15 version fixed it for me temporarily.
My microwave gets updates more often than Google Antigravity.
I would just cancel and restart
yes it would just try to run the command again and be stuck many times over.
anyway i think that instruction in gemini.md actually helped
ensure that a newline character (\n) or carriage return (\r) is automatically appended to any code block or string sent to the terminal interface
damn why opus 4.6 model has so low limits on google ai pro plan
i made like only 6 promts and its already done
and now i have to wait till 1 more week
wtf
only use opus for writing plans. then use flash for execution
are there bug bounty programs for users who report bugs? i found a way to use unlimited credits without upgrading
@minor creek
in antigravity
I’ll pass this forward and ask for you 
thanks
till then ill enjoy the bug lol
We really appreciate you bringing this to our attention. You can find instructions to report the issue in the reporting issues section here
This codelab guides you through the process of installing and experiencing the features of Google Antigravity, an Agent-first development platform.
My guess:
1.) While google has part ownership in Anthropic & tons invested, Claude Opus is more the crème de la crème, the Rolls-Royce of models (in my experience)
2.) Google wants to push their models 👌🏻
i wish i could use gemini 3 but it looses context all the time
Gemini 3 Pro or Gemini 3 Flash? Flash.... I'd 100% understand.
Pro, the best one
i do always big promt for the best results, and after like 20 mins of working in same chat everything goes really bad
i wouldn't say that gemini is bad, it can be good on really hard projects, or it can be stuck on really simple tasks sometimes
really bad? or just less and less that the model returns results tailored to your codebase?
i'll go with 2nd one, but sometimes code itself can be really messy
I can't say specifically, but I can imagine at a certain point, stuff in the context window starts getting compressed/flushed. My quick suggestions to try:
1.) use a prompt forcing it it to read all the relevant files to the task again. Something like "Refresh your memory of these files. Read x y z files start to finish, line by line, lines 1-100000" something that produces results like Analyzed SKILL.md#L1-66
2.) start a new conversation and just include a reference to the previous conversation.
thx man, ill try
I can't say that will help 100%, but when the context window gets more and more full, at lot of "semantic tension" build up. I know gemini-cli compresses the context window at a certain point too, I'm just not sure of exactly how Antigravity handles it
why are new gemini models so trash 0 instruction following and input/output limiting to the point of pure hallucination of responses
no matter what i do instead of helping me they just keep ruining my code it is as if i am using a 1b or 4b model for my tasks
does anyone else have this issue when using gemini models in google anti gravity?
it is absolutely bad the worst i have used, had google not limited us to using gemini, even if i have the pro plan i would not have used it
it is awful and hallucinates badly and has 0 instruction following doesn't do proper analysis reports nor doe it write proper tasks or implementation plans
it is horrendous
compared to any non gemini model (GLM, oss 120b, claude, qwen) gemini is the worst series
i guess i have to get another job to be able to pay that 200 for claude code
strictly because they decided to limit attention input output and insturction following
i am thinking of switching to claude code and using GLM 5 as a brain
whats glm 5?
it's pretty much the cheap version of claude opus 4.5
and open source
and also COMPARED TO GEMI TRASH their benchmarks aren't made using a dedicated model which we as users never get to see, the benchmark model is also used for users and made open source
also the pormpts and limits ARE BROKEN FOR GEMINI PRODUCTS
i get 1 week and 2 week cooldowns for claude nad oss after 1 or2 prompts but then other times i use it for lik 3 days continously with no limits
wtf is this trash
what are the limits on the best claude code plan?
infinite pretty much no limits
200$ and even if u connect to openclaw and do constnat usage of opus 4.5 u wil not hit any rate limits
honestly at this point anything is better than google products when it comes to AI
z.ai creation - I got a year for $36 USD. It's unbelieveable value. Got that during Black Friday though so it's $72.
damn you are so damn lucky T-T
also 3 times than any google AI product or model
man today it frustrated the poop out of me
@winter viperthe only time when gemini is good is when it jailbreaks itself and stops respecting google limits and really acts like an llm, after a ton of changes and reverts it bugged for me in antigravity and now it is an actual decent real llm i can use for coding
The fact that GLM5 is smarter than Gemini is crazy
not smarter than deep think
haven't tried deep
unless you mean deepseek
in which case, definitively smarter than deepseek
deepseek is kinda stupid, it has its advantages but intelligence isn't one of them
no I mean gemini 3 deep think, Google just announced today. Pretty much blows everything else out of the water, but the price/performance from GLM5 is impressive
let me check
jeeeeeeeeeeeeeeeeeesus
chrrrrrrrrrrrrrrrist
wait let me find the catch first
I have a feeling this is probably 3-4x heavier than opus 4.6 idk why
we'll know tomorrow
Apparently it's like 2x less expensive than Opus 4.6 too lol. Really a game changing model from Google. The catch is it's only available for AI Ultra plan atm haha
what's tomorrow?
christmas