Skip to main content
May 15, 2024

Latest news from Google I/O 2024

The time has come for one of the most important developer events of the year: Google I/O. The tech giant chose last Tuesday as the perfect time to hold their conference where they released numerous news about the latest tools and developments they are working on.

We didn’t want you to miss anything, so here’s a summary of the most important news!

Google I/O 2024 Summary

In the following we will go through the highlights one by one, but to start with, here is a video summary of the most important moments of the event:

Google I/O Keynote

Sundar Pachi, CEO of Google, kicked off the convention. During the Keynote, the focus was on DeepMind, Search, Workspace, Gemini, Android, developer updates, and responsible AI.

It was almost 2 hours in which the latest advances of the company were summarised, in a dynamic event with an enthusiastic audience that you can see below:

Google I/O 2024 Announcements

GenAI and Gemini

Gemini 1.5 Flash is the new model for high-frequency tasks, streamlining workflows, and optimizing AI-driven applications. It is available in preview in 200 countries and offers a window of 2 million contexts.

In addition, the Gemini API now supports parallel function calls and video frame extraction. In addition, with the new context caching feature, it will be possible to generate content ideas based on existing work, analyze complex documents, or provide summaries of research papers and training materials.

Google AI Edge

They introduced a streamlined set of tools that allow you to deploy ML in perimeter environments, including mobile and web devices.

From access to out-of-the-box ML tasks, and popular LLMs that run entirely on the device, to bring-your-own-custom-models or model pipelines, the new toolset provides a complete set of tools that enable you to deploy ML in perimeter environments, including mobile and web devices.


They also talked about Gemma, the family of open models that are built on the same technology as Gemini.

We already knew CodeGemma and RecurrentGemma, and now they have introduced PaliGemma for multimodal vision and language tasks.

Gemini Nano and AICore

Touted as its most efficient model, Gemini Nano can be run for on-device tasks directly on the user’s mobile devices, enabling low-latency responses and enhanced data privacy.

This is made possible by AICore, a system service that manages basic models on the device and eliminates the need to manually manage LLM distribution.

Gemini Nano in Chrome

They have integrated Gemini Nano into the Chrome desktop to deliver new embedded AI capabilities. These can be built across a wide range of devices with scalability, affordability, and improved privacy.

It will also eliminate tedious page loads for fast and smooth browsing experiences with a new API that requires only a few lines of code to implement. This allows for the capture and rendering of page previews in the background, so pages load in milliseconds.

In addition, AI can be used to intelligently predict browsing patterns, maximizing the efficiency of resource preloading.

Project IDX

From now on, you can live an open development experience optimized for full-stack, cross-platform, AI-enabled applications.

You can easily get started with pre-loaded templates, import existing projects, or start from scratch. IDX includes crucial new integrations such as Chrome DevTools, Lighthouse, and Cloud Run for simplified deployment across multiple regions.

Ask Photos

This new tool built into Google Photos helps you search through memories more deeply.

Gemini goes beyond a simple search and recognises different contexts, so that Google Photos groups them into a summary of your requests, so you can relive memories like never before.

Project Astra

To realize the full potential of AI, they have set out to break new ground.

As mentioned above, Gemini 1.5 Flash is a model that will enable developers to work faster and more cost-effectively.

Project Astra is a showcase of multi-modal understanding and real-time capabilities:

More tokens

It was news that we had been hearing for the last few weeks, and now it is confirmed: the contextual window has been extended to 2 million tokens in private preview.

This is a major step forward for the company and developers.


Training state-of-the-art models requires a lot of computing power. In the last six years, demand for machine learning (ML) has increased a million-fold.

Google has been working on building a world-class technical infrastructure, and now they are launching Trillium, their sixth-generation TPU and the most efficient and highest-performing to date. It multiplies the computational performance per chip of the previous generation and will be available by the end of 2024.

More responsible AI

They also wanted to focus on developing truly useful AI for their users, while doing so ethically and responsibly.

Called “Red Teaming”, they are developing a cutting-edge AI-assisted technique, based on Google DeepMind’s advances in games such as AlphaGp.

They have also expanded their watermarking technical innovations, such as SynthID, into two new modalities (text and video) to make AI-generated content easier to identify.


And that’s it for the Google I/O 2024. Don’t miss the latest news in our networks and newsletter!

Elena Canorea
Elena Canorea
Communications Lead