Has anyone encountered antigravity gemini model 3.8 being very slow
i updated today to Antigravity IDE 2.6.0 and it has gotten much slower since then
me too, 4 hours trying Antigravity IDE, Antigravity, Antigravity in Android studio, and there is no response
yup for me too , the flash model is 5 to 10 x slower from yesterday
Hi @Muskan2003, @Nasreddine_Elghozi, and @19_DAYANIDHI_CHAUBEY,
Welcome to the Forum.
Since you are already on the latest Antigravity IDE (v2.6.0) release, please try the following to see if response speed improves:
- Adjust thinking level or switch models: From the model selector menu under the prompt box, set the thinking level to
Lowfor faster responses on quick edits, or temporarily switch to another model available on your plan (such as Gemini 4 Argon, Claude, or GPT-OSS-120b). - Start a fresh or compacted session: Start a new conversation (or run
/compactin long threads) so accumulated context from earlier turns does not slow down responses.
If the slowdown persists after trying these steps, please reply here with your operating system and let us know if the delay also happens on a simple prompt in a brand-new conversation.
You’re kidding, right?
After updating from Antigravity IDE 2.55 > 2.6.0 (windows 11) I noticed huge slowdown and massive token consumption (mostly using gemini 3.8 flash high, but i did try Argon briefly before it ate all my tokens). I looked at what antigravity was doing and found it reading my whole codebase, thousands of lines of code across multiple files before doing anything which was consuming all my token allowance and causing huge context bloat, and ended up compacting before even starting to modify code (2.55 seemed to just do efficient targeted reads where it needed to). I also found that when antigravity was running some tests in the terminal it was polling the terminal every few seconds to see if it was alive, and this was also causing more slowdown and token use. A bit of investigation seems to indicate the tools antigravity used in 2.55 for file searching etc have been deprecated in favor of much less efficient methods, and they also added this inefficient polling method to check if background process got stuck. You can re-enable the old tools and try to stop polling via gemini.md instructions and tweaking a json file but we shouldn’t have to do this, google should have tested thoroughly before releasing a version that causes huge degradation in performance. Your reasons for performance degradation may be different to mine. After identifying the issues myself, I asked antigravity IDE to investigate the issues further and it confirmed the issues I’d found, so antigravity does have some ability to diagnose its own inefficiencies and flaws.
Honestly, I’ve found a solution to this. Posted new topic, check it.