Not able to generate commit message in source control

Error generating commit message: [unavailable] error grabbing LLM response: UNAVAILABLE (code 503): No capacity available for model gemini-3.5-flash-lite on the server (error ID: ab860d62eada414cbfa0a15baa9954de)

Hi @Rushikesh_Patel ,

Could you please share more information about the issue you are facing? and also if possible could you please share a reproduction steps to reproduce this issue.

Error generating commit message: [unavailable] error grabbing LLM response: UNAVAILABLE (code 503): No capacity available for model gemini-3.5-flash-lite on the server (error ID: b0df894e59c3488e845fe9955733cd47)
Hi i have this error generating an auto commit at google antigravity IDE.

Same here. It probably hasn’t been working for two weeks now.

Same issue here, and I’ve narrowed it down to the network: it fails only from my ISP, 100% of the time since 2026-09-15.

  • Error: UNAVAILABLE (code 503): No capacity available for model gemini-3.5-flash-lite when generating commit messages.
  • Error IDs: 63ef1d9407d94727ab8675dc116fefb1, 0ea1dc4a7f92433f87635bac70616bdd
  • Reproduces with a clean profile (–user-data-dir, --disable-extensions) and with two different Google accounts.
  • Fails from AS263811 (Cooperativa UniĂłn del Sud, Argentina), with two different dedicated public IPs: 138.204.160./22 (not CGNAT, confirmed by the ISP).
  • Works from the same machine and account via AS11664 (Claro Argentina mobile).
  • The endpoint daily-cloudcode-pa.googleapis.com is reachable from both networks; only the model request is rejected.

It looks like capacity is being denied per IP range/ASN rather than per user. Could you check whether AS2638**/ 138.204.16** is being throttled or routed to an exhausted pool?

Hi @Rushikesh_Patel, @jicasada, and @EvgeniyGuba,

The reason you are seeing UNAVAILABLE (code 503): No capacity available for model gemini-3.5-flash-lite—even when your chat agent is working perfectly—comes down to how the Source Control panel is wired internally:

1. The Root Cause: Hardcoded Model Routing

In Antigravity IDE, the Source Control “Generate Commit Message” button is hardcoded to request gemini-3.5-flash-lite.

  • Even if you have selected a flagship model (such as Gemini 3.8 Flash High or Pro) in your main workspace, the Git UI widget bypasses your session settings and dispatches a lightweight completion request specifically to the gemini-3.5-flash-lite pool.
  • @jicasada’s observation regarding ISP routing is spot-on: Google Cloud routes API requests based on regional egress IP. If the regional serving cluster assigned to your ISP’s route has exhausted its gemini-3.5-flash-lite quota, the endpoint fails with a 503.

2. Immediate Workarounds (Keep Working Today)

While we wait for Google to rebalance regional cluster capacity, here are two ways to generate commit messages without hitting the 503:

Workaround A: Prompt Your Chat Agent Directly (Bypasses Flash-Lite)

Instead of clicking the sparkle icon in the Source Control input box:

  1. Stage your changes in Git (git add -A).

  2. In your active Antigravity chat, simply type:

    “Generate a concise conventional commit message based on my staged changes.”

  3. Because the conversational agent uses your active session model (which has ample capacity), it will inspect git diff --staged and produce the commit message in seconds with zero 503 errors.

Workaround B: Regional Egress Shift (VPN / Hotspot)

As confirmed by @jicasada, switching your network connection (e.g. connecting via a VPN node in another region or using a mobile 4G/5G tether) routes your API handshake to a different Google Cloud edge cluster where gemini-3.5-flash-lite has available capacity.


3. Feedback for the Antigravity Engineering Team:

To make the Source Control panel resilient to regional capacity spikes:

  1. Automatic Fallback on 503: If gemini-3.5-flash-lite returns an HTTP 503, the IDE should automatically fall back to the user’s active workspace model rather than throwing an unhandled UI error.
  2. Configurable Model Setting: Expose a setting in settings.json (e.g. "antigravity.sourceControl.model": "inherit" | "flash" | "flash-lite") so developers can control which model generates their commit summaries.