 |
CEO Sundar Pichai speaking at Google I/O 2026. Photo: Android Central . |
At the I/O 2026 conference held in the early hours of May 20th (Vietnam time), Google announced the latest version of Gemini along with an AI model that simulates the physical world called Omni.
According to CNET , the event took place as Google raced to keep pace with the development of AI models while providing more services to its massive user base.
I/O 2026 is also a highlight as the market focuses on the soaring valuations of OpenAI and Anthropic, both of which are expected to launch initial public offerings (IPOs) this year.
Gemini 3.5 Flash: Fast and Cost-Effective
Google's focus remains on the Gemini ecosystem. At the event, the company introduced Gemini 3.5 Flash, a streamlined version that remains effective while costing half, or even one-third, the price of comparable, more advanced models.
Speaking to reporters before the event, CEO Sundar Pichai shared that Gemini 3.5 Flash is "amazingly fast." Google stated that this will be the default model in Gemini and Google Search's AI mode globally.
"You no longer have to compromise quality for speed," Google's announcement emphasized. The company stated that Gemini 3.5 Flash's cyberattack protection is improved, reducing the rate of malicious content creation and mistakenly rejecting normal commands.
 |
Gemini 3.5 Flash is touted for its high efficiency and fast output speed. Image: Google . |
Gemini 3.5 Flash is optimized for complex agent-based workflows. According to Google, it outperforms Gemini 3.1 Pro on several benchmarks, such as Terminal-Bench 2.1 (76.2% vs. 70.3%) and MCP Atlas (83.6% vs. 78.2%).
When considering the number of tokens output per second, Gemini 3.5 Flash is four times faster than some other newer models. According to Google, the model can shorten the completion time of some tasks that previously took days or weeks at a lower cost, making it ideal for programmers, businesses, scientists , and general users alike.
"To put it into perspective, large companies are processing around 1 trillion tokens per day. If they migrated 80% of that workload to 3.5 Flash, they would save over $1 billion per year, a substantial amount of capital that could be directly reinvested in the business," Pichai emphasized.
Following Gemini 3.5 Flash, Google revealed that Gemini 3.5 Pro is under development and being used internally, with a wide release expected in June.
Gemini Expansion
The Gemini app has also been redesigned with a Neural Expressive language. A new feature is Gemini Spark, a continuously active personal AI agent. This feature runs on the Gemini 3.5 platform and can connect directly with Gmail, Google Docs, or Slides.
What sets Gemini Spark apart is its ability to run in the background, allowing it to operate even when the user locks their phone or computer.
Some uses of Spark include analyzing monthly credit card statements to alert you to subscription fees or hidden charges, checking emails to extract and summarize important information, summarizing meetings and writing them in Google Docs, and more.
 |
New features coming soon to the Gemini app. Photo: Google . |
On Android, users can track live updates and task progress from actors like Spark through a new interface called Android Halo, expected to launch later this year.
By the end of this summer, Spark will also run directly within Chrome, turning the browser into an agent that assists users on the internet.
To ensure privacy, users can opt out of activating Gemini Spark and choose which apps are connected. This feature is expected to be rolled out for Google AI Ultra subscribers in the US starting in late May.
Daily Brief is also a new feature on Gemini, which can provide a personalized summary every morning. Information is compiled from emails and calendars to create a concise report.
 |
The Daily Brief summary feature in the Gemini app. Photo: Google . |
Users can request priority information types and even guide the AI on the next processing steps. Currently, Daily Brief only supports AI Plus, Pro, and Ultra subscriptions in the US market.
Google also revealed several updates to the Gemini app on macOS, including Gemini Spark integration for processing and automating workflows.
Rambler's speech recognition system is also supported on macOS, capable of removing extraneous words to convert speech to text more smoothly.
Create videos with Gemini Omni
Gemini Omni is a model that simulates the physical world. DeepMind has previously developed several similar models, but only for use in robotics and gaming.
With Omni, users can combine images, audio, and text to create high-quality videos, based on Gemini's real-world knowledge base.
According to Google, Omni can accurately apply the laws of physics to the context, such as gravity, kinetic energy, and fluid dynamics. Thanks to its existing knowledge from Gemini, the model is capable of predicting subsequent events.
 |
A video created by Gemini Omni. Photo: Google . |
The model supports natural language editing, and the context remains consistent with each edit. Through commands, users can easily add characters, insert objects, or change the scene.
Omni also supports creating virtual avatars with realistic appearances and voices. All videos have a SynthID watermark to identify content created by AI.
Google says the first version, called Gemini Omni Flash, will be implemented in the Gemini, Flow, and YouTube Shorts apps. In the future, this model will support both video and audio output.
 |
Google emphasizes Gemini Omni's ability to accurately simulate the physical world. Photo: Google . |
Regarding Google, its increasing investment costs have led Wall Street to have high expectations. According to CNET , investors want to see the company deeply integrate AI into its core products to generate sustainable profits. Agent tools could play a crucial role.
Intelligent agents like Spark and Daily Brief are Google's answer. Through a combination of cost-effective Gemini 3.5 Flash and Omni, the company is striving to maintain its position against the competition.
It's not just Google; expectations for AI companies continue to rise, especially after Anthropic announced Mythos. This model is touted as being so powerful that it can detect thousands of unprecedented security vulnerabilities in global software infrastructure.
Source: https://znews.vn/gemini-vuot-khoi-ranh-gioi-chatbot-post1652959.html
Comment (0)