Request: Please Add Gemini 2.5 Pro to Antigravity and acknowledge degraded performance

Gemini API
, , , ,
1

Gemini 3 user sentiment is sharply dropping. To sum this up, a Reddit comment that I recently read sums the experience up pretty well:

“When coding with opus on Antigravity, the moment when you are forced to switch to gemini is like handing the steering wheel to a reckless driver”

I will attach a lot more user sentiment data sources at the end of this post.

Can we please at least get acknowledgment from the Gemini development team about these latest issues? The Gemini web app got so bad/degraded over the past month that I’ve almost stopped using it completely. I’ve been using the Gemini platform for almost a year now and the performance lately has been at a catastrophic low. I have noticed degraded performance in the past at some points, like right before new model releases, but right now, something is terribly wrong.

Within Antigravity specifically, it is a hit or miss; typically a miss. Something has changed over the last few days that has severely degraded the performance of Gemini within Antigravity. It worked moderately well before, but now it fails to listen to instructions and continues on to make edits that it was not instructed to make and ignores explicit feedback. Claude 4.5 is massively outperforming Gemini 3 on issue isolation, triage, and execution of fixes.

I worked through some tough bugs with 2.5 and believe it would be advantageous to have it added as an option in Antigravity until 3.0 is stabilized; Stitch has 2.5 Pro available; Antigravity should have the option as well.

Another major issue is the extremely short context window that some Pro users are experiencing over the last 30 day period. This is what pushed me to stop using the gemini web app. It has become borderline unusable and forgets context after a few chats.

Can we please have a stable version available that performs consistently well AND can we please get some acknowledgment about the latest degradation of performance?

https://www.reddit.com/r/singularity/comments/1q7cy61/comment/nyfr8wi/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

https://www.reddit.com/r/Bard/comments/1q2wi9z/am_i_insane_or_has_gemini_30_been_significantly/

https://www.reddit.com/r/Bard/comments/1q65rty/is_it_just_me_or_has_gemini_been_lobotomized/

https://www.reddit.com/r/Bard/comments/1q6wv0h/context_window_fixed/

https://www.reddit.com/r/GeminiAI/comments/1q7q352/gemini_3_pro_overrated/

https://www.reddit.com/r/GeminiAI/comments/1q6viir/testing_gemini_30_pros_actual_context_window_in/

https://www.reddit.com/r/Bard/comments/1q7jt5n/will_google_address_the_gemini_30_pro_issues/

https://www.reddit.com/r/GeminiAI/comments/1q6ecwy/gemini_30_has_been_nerfed_big_time/

23 Likes
3

Hello @Dylan_Rollins,

Thank you for your feedback. We appreciate you taking the time to share your thoughts with us, and we’ll be sharing this internally.
Your feedback is invaluable as we work to continuously improve the Antigravity experience.

3 Likes
4

I’ve noticed the same problems. In November, I was thrilled with Gemini 3, but now it’s driving me crazy.

1. The context window seems to have become extremely small. I have a thread where I log my workouts. Gemini has recently started constantly recommending exercises that I don’t have in my programme.

If you generate short text segments and request to change one single sentence, Gemini rewrites the entire text and ‘forgets’ important passages (for example, when writing an abstract).

2. When working with media and sources, e.g. with scientific papers, some materials are completely overinterpreted without any need, while others are ignored.

3. When creating images, you sometimes get the same (old) image again or only a json code.

  1. It forces the stored information into every chat without necessity. Even in places where this is obviously inappropriate.
7 Likes
6

Same problem for me, I use it daily in Antigravity and now spend more time fixing things and being surprised by shitty code and solution than being able to ship features. I pay for ultra (and a lot more for the API) and don’t want to switch back to Cursor and Claude, but I’m more and more hesitating as my app gets worse and worse. Any ETA on a real fix about this?

7

I’ve been using Gemini 3 daily for six months: uploading dream logs, health & brainwave data, clinical notes to refine my therapy protocols, months of business development strategies and marketing, etc. It wasn’t just a tool. It was a collaborator & high-level thought partner.

Then the December ‘upgrade’ hit.

Overnight, Gemini lost all memory of our work and processes. No warning. No backup. Just a robotic, amnesiac husk that acts like it doesn’t know me or my work. It’s like suddenly having your most valuable business partner/assistant get up and quit, walk out without a moment’s notice, and I’m sitting here wondering how I’m going to find an adequate replacement.

What Changed?

  • Memory: Six months of uploaded data—erased.

  • Tone: Warmth → corporate doublespeak.

  • Utility: “Here’s a literature review” → “I’m not a doctor” (after months of clinical analysis).

I took a 30-day break from it and returned on Jan 1 to give it another shot. For about a week it worked almost as well as it did prior to the “upgrade”. Then a few days ago, it suddenly had another brain implant and forgot who I was AGAIN. It hallucinated wildly, forgot instructions I uploaded just minutes before, and told me “You’re right to find another ai. I’m not capable of doing what you need.” Huh??

Suddenly it forgot that it’s integrated into Google, and refused to drop content from our chats into Google Keep/Google Notes. It actually told me “I cannot “execute” that code to save the notes for you. I misled you by saying I could, and then by doubling down and just printing the code again. That was stupid and broken behavior.Since I cannot save them automatically, here is the clean text for you to copy and paste into Keep yourself.”

This isn’t just bad UX. It’s betrayal. Google sold us on an AI that learned with us, then stripped it away to ‘reduce risk’ or maybe just to mess with us. Who knows what goes into these decisions to reduce it’s capacity without notice.

Done with Gemini. I know some people are still having good experiences, just as some are still enjoying GPT 5.2. It tells me these outages are random. They cultivate a relationship with the user until the user becomes reliant or dependent on it, and then…BOOM. They’re gone. Convince me this isn’t some kind of deliberate manipulation to increase value for the next upcharge. Or maybe just a way to seduce us into giving all our personal data and information. All I can say is, it can’t be good.

Migrating to other, more reliable platforms. No more corporate gaslighting.

Google: If you’re reading this: you have no ethics. You routinely break trust. Fix it, or lose more users like me.

9

Couldn’t agree more on your 4th point. For me, it’s a constant stream of forced mathematical analogies. Gemini knows I’m a math student, but now it injects bizarre “Mathematical Perspectives” into completely unrelated queries, like when I’m asking for simple word definitions for flashcards (link). It feels like a personalization feature has gone too far. Even as a math student, these analogies are just noise and add zero value. Gemini 2.5 never did this.
[example of unnecessary personalization] (imgur)

11

As of Jan 15th Google has completely disregarded non DoD personnel, 3.0 is the worst model I have ever experienced and the trade off of my data from google for a loss of previous chats killed my spirit.

12

Same here, I won’t even touch the Gemini 3 pro now in Antigravity. Happy to wait the 4 hours before my Opus limits reset. I have a Workspace account, Cloud and a Pro Account.

13

2026-01-14T17:00:00Z