The pro tier works again on agy 1.4, and plugin-menu options now reach the scripts
3 hours ago
If delegating on the pro tier has been failing with exit 14, version 0.29 fixes it. Since Antigravity CLI 1.4, the default pro model is rejected on every plan, so the review command, which uses the pro tier, failed out of the box. This release also makes your saved plugin options work, and rolls up everything since 0.25.2.
Ask about this presentation
Chapters
Show transcriptHide transcript
If delegating on the pro tier has been failing with exit 14, version 0.29 fixes it. Since Antigravity CLI 1.4, the default pro model is rejected on every plan, so the review command, which uses the pro tier, failed out of the box. This release also makes your saved plugin options work, and rolls up everything since 0.25.2.
The pro tier asked for Gemini 3.1 Pro, High. Consumer Google AI plans now list Gemini 4 Argon in its place, and enterprise plans renamed the model to plain Gemini 3.1 Pro. The name the plugin asked for matched neither one. So the review command stopped with exit 14, model unavailable.
The pro tier now defaults to Gemini 4 Argon, High, a name consumer and enterprise plans share. If your plan doesn't serve it and you haven't remapped the tier, the delegate script retries with Gemini 3.1 Pro, then Gemini 3.1 Pro High, and the usage line names the model that ran. It retries only on the immediate "model not recognized" error, so a timeout still exits 12.
Options you set in the plugin menu, like tier_pro, default model and usage log, were silently ignored. Claude Code passes plugin options to hooks, but never to Bash, which is where these scripts run. A new shared loader reads them from your Claude settings file, and an environment variable you set yourself still wins.
If you once saved a pro model in the plugin menu, this is for you. That value used to be ignored, and the default ran. Now it's honoured, and because a remap never falls back, a stale name stops with exit 14 instead of quietly running the default. After upgrading, run the doctor script. It flags a bad name and suggests valid ones.
The doctor got stricter too. It used to pass a tier model whenever its name was a substring of a line in the models list. So Gemini 4 Argon, missing its effort level, and Gemini 3.1 Pro High both read green while the CLI rejected them. Now it matches the exact name, and a miss suggests up to three real candidates.
Cost tracking got easier. Since 0.28, the usage line names the model that ran, its tier, the run's duration in seconds, and its number of turns. In 0.29, a usage log pointed at a missing folder creates it, where before every line was silently dropped. And cost compare prices the pro tier from the model that actually ran.
Running out of Google AI credits or hitting a spend cap now exits 10, an expired print timeout exits 12 even with a partial reply, and an unknown job id exits 1. The migrate tool runs on Windows, the flash tiers moved to Gemini 3.8 Flash, and the price file covers Opus 5.5, Sonnet 5.5 and Gemini 4 Argon.
To upgrade, run these two commands inside Claude Code. If you saved a pro model, check it with the doctor right after. One limit: on a plan with none of the three pro models, the fallback costs up to three quick failed starts before exit 14. Everywhere else, the pro tier works again on Antigravity CLI 1.4.









