Google Cloud Announces Gemini Agent Preview, Claims One-Third Cost Reduction via Model Routing
Google Cloud launches Gemini Agent enterprise preview with four-tier memory and cross-model routing; vendor-reported cost and speed figures lack disclosed methodology.
ImportanceMaterialEvidenceE2 unreplicatedWrite-upQuick
Google Cloud announced Gemini Agent, an enterprise general-purpose agent preview, at its Gemini at Work 2026 conference on October 9. The platform is designed to autonomously decompose tasks, invoke tools, and sustain long-running work in the cloud without user supervision at each step.
The product introduces four memory tiers — session, semantic, procedural, and episodic — and supports intelligent routing across models including Gemini and Anthropic's Claude. Google Cloud claims this routing reduces comprehensive operating costs to one-third of using frontier models exclusively, while improving execution speed by 40%.
As reported by Zhidx, the company did not disclose the test task set or baseline models behind those figures. Gemini Agent is currently available only as a limited preview with no announced commercial launch date or pricing.