Google switches Gemini to compute-based limits, cuts AI Ultra to $100/month

Google announced at I/O 2026 that the Gemini app is moving from daily prompt limits to a compute-based model that measures actual resource usage. The new system factors in prompt complexity, features used, and chat length—replacing the previous blanket daily limits.

Limits now refresh every five hours until users reach their weekly cap. According to Google, when users hit their limit on larger models, the system automatically shifts them to "state-of-the-art, lightning-fast smaller models" to maintain continuous service.

Google says the change provides "a better way to allocate limits, because a simple text prompt uses far less compute than a complex video or coding prompt." The Gemini app will soon offer pay-as-you-go AI credits for additional compute—a feature already available in Google Antigravity and Flow products.

AI Ultra pricing restructure

AI Ultra now starts at $100 per month, positioning it for "developers, technical leads, knowledge workers, and advanced creators." The tier includes:

• 5x higher usage limits in Gemini app and Google Antigravity versus Pro plan
• Gemini 3.5 Flash integration for testing and debugging
• Priority access to Google Antigravity agent development platform
• 20TB cloud storage
• YouTube Premium individual plan
• Access to Gemini Spark beta (rolling out to US subscribers next week)

The previous $250/month tier drops to $200 with unchanged capabilities, including 20x higher usage limits than Pro and access to Project Genie with Street View integration.

What this means

The shift to compute-based metering aligns Google with industry trends toward usage-based pricing, similar to token-based billing from OpenAI and Anthropic. This change could benefit power users who make simple queries while penalizing those running complex multimodal or coding tasks. The AI Ultra price cut to $100 suggests Google is competing more aggressively for developer mindshare against GitHub Copilot ($10-20/month) and Cursor ($20/month), though at a significantly higher price point that bundles cloud storage and YouTube Premium. The automatic downgrade to smaller models when hitting limits is a notable UX decision—maintaining service continuity but potentially degrading output quality without explicit user consent.