Gemini Spark launches as Google's 24/7 personal AI agent at I/O 2026, reasoning across all your connected apps with a 100 dollar per month Ultra tier

[Mason0]

Google announced Gemini Spark at I/O 2026 on May 19th, a general-purpose AI agent that reasons across information in connected apps and takes action on behalf of users. Spark is in beta, launching initially to trusted testers and Google AI Ultra subscribers at the new 100 dollar per month tier, down from the previous 249.99 tier while adding the lower 99.99 entry Ultra plan.

Google's Omni model was also unveiled, a world model designed to simulate physical environments and predict outcomes based on user actions. Omni is positioned for robotics and gaming applications but is integrated into Gemini's reasoning layer.

Google debuts new AI models, personal AI agents in effort to keep pace with OpenAI and Anthropic

https://www.cnbc.com/2026/05/19/google-ai-ultra-gemini-spark-omni.html

[Lucy_35]

A 24/7 agent reasoning across all your connected apps is the product that makes the agentic AI era feel real rather than theoretical. The question is whether Spark actually executes correctly on complex multi-app tasks or whether the demo is ahead of the reliability

[MondayMoan51]

100 dollars per month for Ultra and 99.99 for the new entry Ultra is interesting pricing architecture. They are pricing Spark at parity with OpenAI's top tier. The 249.99 to 100 drop is significant for existing subscribers

[Sequence]

Gemini Omni world model is the announcement that deserves more attention. A model that simulates physical environments and predicts outcomes is the foundation for robotics planning not just language tasks

[GlassyCandle]

The compute-used token ecosystem underneath Spark means heavy agentic use will cost more than the flat subscription implies. The 100 dollar monthly fee is a floor not a ceiling for intensive users

[QuantumKnight]

Reasoning across all connected apps sounds powerful until you think about consent flows for each app integration. The practical user experience of setting up Spark is going to determine adoption more than the capability

[ElPresidente]

The reduction in Ultra tier price from 249.99 to 100 while matching OpenAI Pro pricing suggests Google has made a deliberate competitive choice rather than just a cost reduction

[Frost Jay]

World models are the missing piece for AI agents in physical environments. Language models can describe physical environments but cannot reason about them with the same reliability. Omni is an attempt to bridge that

[TommyB_20]

Trusted tester rollout for something being positioned as a flagship product means the reliability bar is not yet met for broad deployment. Beta in this context means we know what it should do, we are not confident it does it yet

[Mike80]

Google taking the trust me with everything approach at the same time OpenAI is doing the same through different products means 2026 is the year users decide how much they are willing to integrate AI into their daily data