One developer using Google's Antigravity coding tool was tracking more than 300 million weekly input tokens in late 2025. By January 2026, the same account was hitting limits at under 9 million — a collapse of more than 97%. Google hadn't announced a policy change. It had restructured its pricing.
The company is now steering professional developers toward its AI Ultra plan at $249.99 per month, framing the existing AI Pro tier ($20/month) as suitable only for "hobbyists, students and developers who live in the IDE." Under the new structure, Pro subscribers are limited to Gemini 3 Flash under a weekly refresh cycle, a significant departure from the previously advertised five-hourly quota reset. Additional credits are purchasable at $25 per 2,500, though Google has not disclosed what a single credit translates to in token consumption — leaving developers unable to forecast costs with any precision.
The opacity isn't incidental. Antigravity launched in November 2025 with deliberately elastic language — "high," "generous," and "meaningful" — to describe its limits. Reports of unexplained "ghost-drains" on quotas and mid-week lockouts have piled up in Google's AI for Developers forum and across Reddit, with developers repeatedly asking for a straightforward token-to-credit conversion table. Google has not provided one.
Antigravity supports five LLMs — Gemini 3.1 Pro (High/Low), Gemini 3 Flash, Anthropic Claude Sonnet 4.6, Claude Opus 4.6, and OpenAI's GPT-OSS 120B — a multi-model lineup that had distinguished it from more narrowly focused rivals. But the pricing shift arrives at an awkward moment: Cursor recently reported crossing one million paying users, and GitHub Copilot has moved toward more flexible per-seat enterprise pricing. Google may have underestimated how quickly its developer base would start comparison-shopping.