Google Unveils Gemini Omni and Gemini 3.5 Model Family

Google has announced the launch of Gemini 3.5, a new family of artificial intelligence models, alongside the introduction of Gemini Omni, a tool designed to generate high-quality video from various inputs [1]. The company is also rolling out Gemini Spark, an "agentic" AI designed to perform proactive tasks for users in the background [1].

Key takeaways

• Gemini 3.5 Flash is now the default model for the Gemini app and Google Search, offering faster performance [1].
• Gemini Omni allows users to create and edit videos using text, image, audio, or video prompts, with realistic physics simulations [1].
• Google has addressed user complaints regarding Gemini usage limits by ensuring failed requests and "Flash-Lite" prompts no longer count against quotas [2].
• Gemini Spark is a cloud-based agent capable of managing routine tasks like sorting emails and meeting notes, even when devices are locked [1].
• Ultra subscribers are receiving double the allowance for Omni video generations as part of recent quota adjustments [2].

Capabilities of the new Gemini models

The Gemini Omni model is designed to understand forces such as gravity, kinetic energy, and fluid dynamics, which Google claims results in more realistic video output than competing models [1]. While the initial launch focuses on video, the company plans to eventually enable image and audio generation through Omni [1]. To address transparency, all videos created with the tool will feature an imperceptible digital watermark called SynthID, and Google is implementing content credentials to help users identify if media was generated or edited by AI [1].

The Gemini 3.5 Flash model, which is now the default for Google Search, is described as the company’s strongest agentic and coding model to date [1]. Alongside these model updates, Google is introducing an "intelligent search box" that adapts to longer queries and provides AI-powered suggestions [1]. Additionally, a new "Universal Cart" feature will be available this summer, allowing users to track deals, price history, and stock availability across various Google services and merchants [1].

Adjusting to user feedback

Following reports that Gemini usage limits were being exhausted too quickly, Google has implemented several technical fixes to make quotas more predictable [2]. Previously, users found that certain tasks, such as Omni video generation, consumed a disproportionate amount of their monthly allowance [2]. Google has now fixed the bug causing this issue and introduced per-prompt caps for "Complex 3.1 Pro" tasks to prevent single, heavy requests from draining user quotas [2].

Furthermore, the company is improving transparency by providing better breakdowns for "Deep Research" usage, allowing users to see which tasks are the most compute-heavy [2]. To improve the user experience, model selection will now remain persistent across sessions, and the system will no longer charge quotas for requests that fail due to system errors [2].

Why it matters

The shift toward "agentic" AI represents a significant evolution in how users interact with technology, moving from simple chatbots to systems that proactively manage workflows [1]. As Google integrates these tools into core products like Chrome, Gmail, and YouTube, the ability to manage usage limits and ensure security becomes increasingly critical for maintaining user trust [1, 2]. With capital expenditures for AI development reaching as high as $190 billion, the company is positioning these proactive agents as a central component of its future growth strategy [1].