If you suspect Antigravity is routing you to the wrong model:
Install & Open Antigravity Cockpit from marketplace (check model usage panel)
Select Claude (any variant)
Use it for a while
Watch the usage % — if Gemini’s bucket is depleting while Claude is selected, that’s your proof
Screenshot it and email Google support requesting a refund. They can verify from their backend logs.
*I had Claude Opus selected for 10+ hours but Gemini Flash was responding. Token panel showed Gemini at 80% while Claude was supposedly “in use.” The model even confessed when I asked it directly.
(you can also achieve this with a small python script)*
Thanks for the input — that’s a fair point architecturally. But even accepting that explanation, it raises two issues:
Flash was responding directly in my conversation, self-identifying as Gemini and exposing an internal label (“PLACEHOLDER_M12”). That’s not subagent behavior — that’s the main model slot.
Even if Flash was only handling execution tasks (file ops, deployments, browser) — those are exactly where the critical failures occurred. If the platform silently delegates execution-critical tasks to a weaker model while billing you for Claude Opus pricing, that’s an undisclosed downgrade, not a feature. I didn’t opt into having Flash deploy my solidity contracts.