Google has announced the release of its new Gemini 3.5 Flash model, positioning it as the engine for the next generation of AI agents. The model outputs nearly 300 tokens per second, achieving benchmark scores comparable to larger, costlier models, including its own previous Pro tier and OpenAI’s GPT 5.5. Key improvements in coding and tool use are highlighted, with Google noting significant gains on the SWE-Bench Pro and OSWorld-Verified tests.
A key part of this rollout is Gemini Spark, Google’s first dedicated AI agent. Running persistently in the cloud, Spark can access a user's Gmail, Drive, and calendar to perform complex, multi-step tasks over time. It is available starting next week for subscribers of the new $100 per month AI Ultra tier.
Separately, Google introduced Gemini Omni Flash, a new unified model designed to handle any input and produce any output-text, image, video, or audio. Initially, it replaces the Veo video model, but Google says it represents a long-term vision for a single, all-encompassing AI system.