I’ve been noticing that long Gemini reasoning turns in Antigravity are consistently interrupted at around the 120-second mark.
When this happens, Antigravity starts a new Thought block automatically. If the model still needs more time, the same thing happens again, so it can end up looking like this:
-
Thought for 120s
-
Thought for 121s
-
Thought for 121s
-
Thought for 121s
-
Thought for 121s
I checked the local transcript.jsonl, and the interruptions line up very closely with 120 seconds:
-
16:17:41 → 16:19:41 — 120s
-
16:20:03 → 16:22:03 — 120s
-
16:22:07 → 16:24:09 — 122s
-
16:24:14 → 16:26:15 — 121s
-
16:26:27 → 16:28:28 — 121s
-
16:28:56 → 16:30:58 — 122s
-
16:32:00 → 16:34:04 — 124s
Each time, the log records:
Error: The stream was interrupted. Please continue the task you were working on.
The main problem is that after the interruption, Gemini does not always resume cleanly. It may repeat part of the previous reasoning, hit the same ~120-second interruption again, and get stuck in a loop.
I’ve seen this happen in both normal agent tasks and subagent tasks, so it does not seem to be specific to subagents.
Screenshot attached showing one occurrence with several consecutive 120–121 second Thought blocks.
Antigravity version: 2.18.1
Model: Gemini 3.1 pro high
