Google's Biggest Gemini AI Updates of 2026: Everything Announced So Far

Google's biggest AI news of the year didn't trickle out over separate announcements; it landed all at once at Google I/O 2026 on May 19–22. In one keynote, Google reshaped its Gemini app, shipped a new flagship-beating model, and introduced two features designed to make Gemini feel less like a chatbot and more like an assistant that works in the background.

Here's what actually happened, and where things stand heading into the second half of the year.

Gemini 3.5 Flash: The "Cheap" Model That Beat Last Year's Flagship

The headline release was Gemini 3.5 Flash, the first model in the new Gemini 3.5 family, which went live on May 19, 2026. What made it notable wasn't just that it shipped; it's that a Flash-tier model, historically the fast-and-cheap option, outperformed the previous Pro model on several difficult benchmarks.

According to Google's own published figures, Gemini 3.5 Flash beat Gemini 3.1 Pro on Terminal-Bench 2.1 (76.2%), MCP Atlas (83.6%), and GDPval-AA (1,656 Elo), while running roughly four times faster in output tokens per second. It also posted strong multimodal results, scoring 84.2% on CharXiv Reasoning.

A few practical details worth knowing:

• It carries a 1-million-token context window with multimodal input (text, images, audio, video, PDFs).
• API pricing is $1.50 per million input tokens and $9 per million output tokens — about three times the previous Flash model, though still cheaper than Pro-tier pricing.
• It's free to use in the Gemini app and in AI Mode in Google Search, where it's now the default model globally.
• It still trails Gemini 3.1 Pro on the hardest academic reasoning benchmarks (like Humanity's Last Exam and ARC-AGI-2), so it isn't a clean 1:1 replacement for Pro on every task.

Gemini 3.5 Pro, the higher-capability model in the family, was notably absent from the May launch. Google confirmed it's still in internal testing, with a public release expected later.

Gemini Spark: A Personal AI Agent That Runs in the Background

Google also introduced Gemini Spark, described as a 24/7 personal AI agent built on Gemini 3.5 Flash. Unlike a chatbot that only responds when prompted, Spark is designed to keep working in the background. It can:

• Organize files
• Assemble documents
• Handle multi-step tasks across Google Workspace under a user's direction, even when the phone is locked.

Spark began rolling out to trusted testers at launch, with a beta planned for Google AI Ultra subscribers shortly after.

Daily Brief: A Morning Digest Pulled From Your Inbox and Calendar

Another new feature, Daily Brief, acts as a personalized morning summary. It pulls information from a user's inbox, calendar, and outstanding tasks, then organizes it into a prioritized overview, surfacing the most important items first rather than just listing everything chronologically. Daily Brief began rolling out to Google AI subscribers in the United States.

A Redesigned Gemini App: "Neural Expressive"

Google also rebuilt the Gemini app's interface from the ground up under a new design language it calls Neural Expressive. The redesign brings:

• Fluid animations
• New typography
• Haptic feedback

It changes how responses are presented, leading with bolded key information at the top, with supporting details, images, and timelines appearing as the user scrolls, rather than a single wall of text.

Gemini Omni: A New AI Video Generation Model

On the creative side, Google introduced Gemini Omni, a new AI video model that combines Gemini with Google's generative media tools. It can take a prompt like "claymation explainer of protein folding" and generate a video. It supports uploading audio, images, or videos as references to maintain visually consistent outputs.

Transparency Tools: SynthID and Content Credentials Expand

Google also expanded its AI-content transparency tools. SynthID, its AI-content watermarking and detection system, is expanding into Search and Chrome alongside Content Credentials. This feature shows whether a piece of content:

• Originated from a camera or an AI model
• Has been edited with generative tools

Google also said OpenAI, Kakao, and ElevenLabs are adopting SynthID, joining Nvidia as cross-industry partners.

Gemini for Science

For research use cases, Google introduced Gemini for Science, which combines Gemini's reasoning with Deep Think and Deep Research capabilities. It connects to more than 30 life-science databases and tools through Google's agentic platform, Antigravity.

By the Numbers

Google shared a few scale metrics at the keynote worth noting for context:

• Monthly token processing across its surfaces reportedly jumped approximately sevenfold year-over-year to more than 3.2 quadrillion tokens per month.
• The company said more than 8.5 million developers are now building on its models monthly.

What This Means

Taken together, these updates show Google leaning into two ideas at once: making Gemini feel more proactive (Spark, Daily Brief) and making its models cheaper and faster without giving up much capability (3.5 Flash). That's a direct response to competitive pressure from OpenAI's ChatGPT and Anthropic's Claude, both of which have introduced agentic and multimodal features this year.

The bigger release is Gemini 3.5 Pro; it's still to come, and it's expected to close the gap that 3.5 Flash currently leaves on the hardest reasoning and long-context tasks. Until then, Gemini 3.5 Flash is Google's default answer across nearly all of its consumer and developer surfaces.