Google I/O 2026: Gemini 3.5 Flash, Spark, and Omni, what’s new this week

Google held its I/O 2026 conference on May 19 and unveiled four announcements directly impacting freelancers, SMEs, and agencies.

No teasing, no vague roadmap: everything is available this week or next.

Here’s what’s changing this week, without the marketing fluff.

The Gemini 3.x family has been around for several months.

What’s new this week: 3.5 Flash, Spark, Omni Flash, and an AI Ultra at half the price.

Gemini 3.5 Flash: the new default engine, faster and cheaper

Gemini 3.5 Flash was released for general availability on May 19, 2026, the day of the keynote.

It is now the default model in the Gemini app, Google Search AI Mode, the Antigravity dev platform, and the Gemini API.

Google’s pitch: 4 times faster than other comparable frontier models, at less than half the price.

On the published benchmarks, 3.5 Flash surpasses Gemini 3.1 Pro in coding (Terminal-Bench 2.1 at 76.2%), agentic (MCP Atlas at 83.6%), and multimodal (CharXiv at 84.2%).

In practical terms: The smaller Flash model now outperforms the larger Pro model of the previous generation, for a fraction of the cost.

Available worldwide, free tier included, no waiting list.

When the “fast and cheap” model outperforms last year’s “premium” model, the entire market shifts downward in price.

The official product page details the complete benchmarks.

Gemini Spark: the personal agent running 24/7 in the cloud

Gemini Spark is the real breakthrough of the I/O 2026 announcements.

It’s no longer a chatbot that responds, but an autonomous agent that acts for you, even with your phone locked.

It runs on dedicated Google Cloud virtual machines and operates continuously, without needing your laptop to be open.

Under the hood: Gemini 3.5 Flash + the Google Antigravity platform to orchestrate multi-step actions.

Real-world use cases demonstrated by Google

Here are the examples highlighted during the keynote, relevant to freelancers or SME managers:

• Monitor bank statements to spot forgotten subscriptions eating into cash flow
• Sort school emails from children and prepare a weekly summary
• Aggregate scattered project notes in Gmail into a single Google Doc
• Compose emails via voice command, with manual validation before sending
• Launch in-depth research that runs overnight

Spark integrates natively with Gmail, Docs, and Workspace, plus partners like Canva, OpenTable, and Instacart.

The positioning is directly against OpenAI’s super app strategy around ChatGPT, aiming to become the single entry point to all digital services.

Support for the MCP protocol for third-party tools will arrive this summer, opening the door to any SaaS that exposes its functions.

Important safeguard: Spark always asks for confirmation before sensitive actions (sending emails, purchases, calendar additions).

Availability timeline

Trusted testers this week, beta for AI Ultra subscribers in the US next week (around May 26, 2026).

The desktop app arrives this summer 2026, with Chrome integration to follow.

For Europe, no confirmed date, but Google’s usual pattern suggests a gradual rollout over the summer and fall.

The official Spark page centralizes demos and the waiting list.

Gemini Omni Flash: the model that creates and edits video

The third major announcement: Gemini Omni Flash, a model specialized in multimodal video creation.

You provide it with text, an image, sound, or video, and it produces or edits a video output.

Its unique feature: It simulates real physics (gravity, collisions, material properties), eliminating major video AI flaws from previous generations.

Available immediately for AI Plus, Pro, and Ultra subscribers in the Gemini app and Google Flow.

To test without a paid subscription, it’s free this week via YouTube Shorts Remix and the YouTube Create app (for users 18 and older).

Every video generated by Omni includes the SynthID watermark, verifiable via the Gemini app or Chrome.

AI Ultra cut by 2.5 and restructuring of paid plans

The Google AI Ultra plan drops from $250 to $100 per month, a 60% decrease.

For $100, you get: 5 times the usage limit on the Gemini app, 20 TB of Google One storage, YouTube Premium included, and beta access to Spark.

The AI Pro plan remains at $20/month, unchanged for freelancers and independents.

A higher tier at $200/month has been created for power users with high consumption.

This restructuring aligns Google with ChatGPT Pro pricing ($200/month) and undercuts challengers positioned in the “$200-250” segment.

The pattern mirrors OpenAI’s pricing shift on Workspace Agents in early May 2026, indicating a gradual alignment of major publishers on “agent-included” plans.

What this means for freelancers, SMEs, or agencies

Three practical implications to remember this week.

For the solo freelancer

The AI Pro plan at $20 remains the best entry point to automate recurring tasks with Gemini 3.5 Flash.

The Spark agent is not accessible at this price, so there’s no need to switch unless there’s a proven need for a persistent agent.

The real immediate benefit: the Gemini app is faster and more accurate, without changing your subscription.

For SMEs

If multiple employees already use generative AI daily, the new Ultra tier at $100 becomes attractive when pooling usage.

Spark can handle repetitive tasks on Gmail/Docs (client recaps, email inbox summaries, quote aggregation) without needing Zapier or Make.

Keep in mind the safeguard: every sensitive action is still manually validated, so there’s no risk of accidentally sending an email to a client.

For agencies and creative studios

Omni Flash opens up new possibilities for video teams: rapid scene prototyping, conversational editing, character consistency between shots.

Combined with Google Flow, the “brief → storyboard → first version” pipeline compresses to a few hours instead of days.

On the client billing side: it’s now possible to offer video variations upfront, which was unthinkable at the hourly rate of an editor until now.

The same logic applies to static design: the question of the role of the designer and front-end developer in 2026 resurfaces with each new generative AI tool.

The real topic of summer 2026: Learning to delegate long tasks to an agent, not just better prompting a chatbot.

FAQ

Is Gemini Spark available in France?

Not yet: The beta starts the week of May 26, 2026, for AI Ultra subscribers in the US, with Europe to follow in waves without an official date announced.

What model does Gemini Spark use?

Spark runs on Gemini 3.5 Flash and the agent platform Google Antigravity.

What’s the difference between Gemini 3.5 Flash and Gemini 3.1 Pro?

3.5 Flash is newer, faster, cheaper, and outperforms 3.1 Pro on coding, agentic, and multimodal benchmarks according to Google.

How much does the new Google AI Ultra plan cost?

$100 per month, down from $250, with 5 times the usage limit, 20 TB of storage, YouTube Premium, and beta Spark access.

Is Gemini Omni Flash available for free?

Yes: Via YouTube Shorts Remix and the YouTube Create app (users 18 and older) starting this week.

Can Spark act without my authorization?

No: The default safeguard requires manual confirmation before any sensitive action like sending an email, making a purchase, or modifying the calendar.

What third-party tools does Spark integrate with?

At launch: Gmail, Docs, Workspace, Canva, OpenTable, and Instacart, with MCP protocol support planned for this summer to open up to all compatible SaaS.

When is Gemini 3.5 Pro coming out?

Google announced a release in June 2026, a month after the I/O keynote.

What is Antigravity 2.0?

Google’s agent-first dev platform, available globally since May 19, 2026, serving as the foundation for Spark and custom agents built via the Gemini API.

Should I switch from ChatGPT or Claude to Gemini?

No: The I/O 2026 announcements make Gemini very competitive in terms of price/performance, but the best approach is to test each model on your own use cases before migrating.