Subject: Gemini 3.8 Flash — 2–5 min latency for trivial prompts, no error shown (reproducible today)
Model: gemini-3.8-flash-low and gemini-3.8-flash-medium (both show the same behavior)
IDE: Antigravity IDE 2.13.0, Windows 11
Account plan: Google AI Pro
Region: Egypt (direct network access, no VPN/proxy)
Repro session: 20261105T143557 (Nov 5, 2026, ~14:35 UTC+2), workspace on a small React project
Symptom:
Every prompt — including a bare “Hello” — takes 2–5 minutes before the
first token arrives. The panel stays in “Working” state with no visible
error. Local machine, auth, and network are healthy (RTT to Google
endpoints ~1s, no proxy).
Local diagnostics:
Language server and auth initialize in under a second; no 503 or retry
lines appear in local logs at default verbosity (2.13 appears to run the
agent cloud-side, so backend failures are silent).
Not local: other models respond in seconds on the same machine and
connection.
Same reports, growing daily:
GitHub issue google-antigravity/antigravity-cli #1246 — repeated 503
“No capacity available for model gemini-3.8-flash-*” with silent retry
backoff (4s … 24s), 2–5 min total; Claude on the same box: 14s vs 159s
Forum thread #198333 opened today (Nov 5): “Critical Performance Issue:
Gemini 3.8 Flash Extremely Slow in Antigravity IDE”
Request: please review 3.8 Flash serving capacity / cluster routing, and
surface the 503 retries in the UI instead of a silent “Working” state.
Currently forced to switch models, which defeats the purpose of the Pro plan.
Since you’ve already tried restarting Antigravity, starting a new conversation, and comparing against other models, this doesn’t look like a local machine or network issue—the 2–5 minute stalls in the “Working” state match transient backend capacity limits (HTTP 503 retries) on Gemini 3.8 Flash.
We’re not able to reproduce those multi-minute delays on Gemini 3.8 Flash right now, which points to temporary capacity contention around November 5 rather than a persistent issue. If you run into this slowdown again, here is what you can do right away:
Open the model selector menu under the conversation prompt box and switch to another available model for your plan (for example, Gemini 4 Argon or Claude Sonnet 5.5 (thinking) on Google AI Pro/Ultra, or another model available in your picker).
If you switch back to Gemini 3.8 Flash for quick questions or simple edits, keep the thinking level set to Low or Medium in the model selector.
You don’t need to reinstall or clear local app data for this, as model capacity is handled server-side. If you’re still hitting 2–5 minute “Working” delays on Gemini 3.8 Flash today, please reply with your installed Antigravity version and any error output or logs from the session so our team can take a closer look.
Hello, I have the same proble, a week after the first report. Gemini Flash 3.8 is totally slow, sometimes it takes 20 minutes to do a simple task. Totally unusable. This started to happen yesterday or day before yesterday yet the situation is worse every hour.
Extremely slow, probably starting from yesterday. Even a simple request asking it to say “hello” could take anywhere from tens of seconds to several minutes.
Utterly slow and unreliable. I can even type hello and the thing wont spit back for ages. It seems to be like this for around 2 weeks or so now. I have very good hour(s) then boom nothing. Honestly its so sad to see google fumble this bad. It even happens on hydrogen …
I am having the same issue as well. My Gemini Pro plan’s limit got reset and found out Gemini flash 3.8 to take unexpectedly too long to even making 1 line (in total) code edits.