Gemini 3.8 Flash is

Gemini 3.8 is so slow and hallucinates so much that it’s impossible to work with. It’s really, really bad. The worst model ever.

How can you treat your customers like this?
I’m starting to think that Google hates developers. Antigravity was such a great piece of software, but every update seems to make it worse.
First, you split Antigravity into separate products, making the whole experience unnecessarily complicated and frustrating.

Now, you’re removing Gemini 3.7 Flash from Google AI Pro subscribers.
Are you guys crazy??

Totally agree with you. They also completely removed Gemini 3.1 Pro for Pro subscribers and locked the new Gemini 4 Argon behind the $100 Ultra tier inside the IDE. I just created a detailed topic about this absolute disaster of an update. Google is literally forcing us to switch to Claude Sonnet.

3.8 flash is one of the best models out there, i dont understand how you can call it slow neither?

I get 200+tps at all times, and sub second ttfb

100% true. Actually I only found the 3.6 as the Balanced model compared to 3.7 and 3.8 with quota usage and the accuracy. 3.8 is very slow and is doing more than what we tell it to do. It even falls into continuous file reading loops.

Now they removed the 3.6 and 3.7 :expressionless_face:

Antigravity was really a good Agent based IDE at first, but with every single update they killed it. They don’t even fix the issues we shared on this forum.

Even people are facing issues with Angravity 2.0 and IDE with opening issues. I feel like they have already killed the product.

They are now removing the Claude models like Claude Sonnet 4.6 (Thinking) from free plans. How many developers would stop using Antigravity for all this stuff.

  • Model hallucination
  • Continuous file reading loop
  • Losing workspace chats memory
  • IDE opening issues with the new Antigravity 2.0.
  • High qouta usage and slowness with models like 3.8
  • Login issues for some
  • Network issues with HTTP errors
  • Terminal character invisibility issues

This is what I experienced so far. How many issues are out there that other developers are experiencing… Antigravity was such as a great product. Let’s hope they could see the reality first instead of chasing the big money.

Fully agree, the 3.8 flash speed is too slow, and I have to click through a bunch of files’ approval commands—it’s really annoying :distorted_face:

Same issue here…

Same, hate to work with 3.8 flash always used 3.7 flash

My suggestion is to try some skills to overcome this problem, maybe this plugin is suitable for you.

Vibesplug

I am moving, no more will I help train their models going forward

Now this is good reason to leave google antigravity.

“Sonnet 5.5 is now available on paid Pro and Ultra plans. Third-party model access will no longer be available on your current plan starting on November 2, 2026.”

Agree.
Don’t know why google discontinue all other moedls like Flash 3.5 Which i love.
3.8 model is high token consumption and very slow in response

I literally just tried using it and ran into a huge problem, it completely bugged out my project. I never had any issues with Gemini 3.1 Pro. Now, with this 3.8 Flash, only beginners claim it’s good; if you use complex code, it’s going to mess up your project. I suspect Google did this on purpose so people would buy the more expensive plan. I miss the days when Antigravity first started.

sabah 9’dan beri projem uzerinde calisiyorum ve hizim 3.8 flash yuzunden bir kaplumbaga gibi cunku cok yavas. kucuk isler yaparak para kazanmaya calisan insanlari zaten parasini verdigi bir hizmetten mahrum edip ayda $100 ode demek zorbaliktan baska bir sey degil. google tam bir hayal kirikligi.

Yeah fr, it’s the only move left bro!

Exactly, only those who use 3.8 Flash as a chatbot would say it’s “the best” model.

It hallucinate a lot, changing my variable names without asking, then on top of that…

It has a very sensitive (hyper-aggressive) filter towards words like “screwing”, “pounding”, etc. which is a very common word to use in creative writing about actual mechanics. I’m screwing the bolt, pounding the nail hard with the hammer.

Yeah, that’s a fair point too. But the issue is, we can only blame and share our opinions here… At the end we have to adjust to what they do.

True. I like 3.5 and 3.6 flash. But 3.8 is slow and not accurate as I expected. It’s very disappointing to see they removed all other models and keep only 3.8…:pensive_face:

my issue is 3.8 Flash searching git commit history to find the code block, and asks for too many confirmation. Previously on 3.6 Flash it searched using normal file system, now it uses Git and many times it doesn’t focus on the task itself but try to resolve side things first then focus on task, It never directly attack the task that is told but always does side tasks first.

That’s because you are using it for basic stuff.

Totally agree with you.

3.8 Flash is a good chatbot, not for doing actual work.