JB Alayrac

You can now sample at higher frame rates (default 1 FPS), and specify start and end times for videos in the Gemini API! We’ve been blown away by all the ways developers are using Gemini to process videos, and see a ton of devs manually clipping and slowing down videos to use…

We just shipped video FPS support in the Gemini API, so you can dynamically customize how many frames per second you want the model to see, unlocking lots of interesting new video use cases! 📹

Pre-training is not dead

The newly generally available Gemini 2.5 Flash and Pro are even better at video understanding than the versions we shared in the blog a month ago, see more details in the tech report 😀

Many Congratulations to @jianyuan_wang, @MinghaoChen23, @n_karaev, Andrea Vedaldi, Christian Rupprecht and @davnov134 for winning the Best Paper Award @CVPR for "VGGT: Visual Geometry Grounded Transformer" 🥇🎉 🙌🙌 #CVPR2025!!!!!!

By popular request, you can now specify frames per second (fps), as well as start and end times, for videos in AI Studio ⏩

People want Veo 3 so we are giving access to Pro subscribers in 71 countries as of...now!

Excited that our work on Gemini Robotics and Gemini spatial understanding have just been featured on #GoogleIO stage! I believe that a frontier model possessing strong real-world understanding capabilities represents the ultimate path to embodied AGI, and we are making rapid…

cooking up something tasty for tomorrow...

The Gemini 2.5 models are magical for analyzing sports video. We asked Gemini to find Draymond's defensive plays from a highlights reel, which requires the model to: - reason “over pixels” to identify defensive plays - identify players in the video using its world knowledge -…

Worth reading this. The video understanding capabilities of Gemini are fantastic.

Gemini 2.5 Pro sets the state of the art on our newly released Minerva video reasoning benchmark by scoring 63.5%. 📜 Paper: arxiv.org/abs/2505.00681… 📊 Dataset: github.com/google-deepmin…

A lot of work went to make Gemini 2.5 SOTA at video understanding, check out this 🧵 for more details! Looking back at where we were a year ago, the progress really feels phenomenal! So many things to unlock and enable from video 🎥 and we are only getting started!

Gemini 2.5 Pro is incredible at video understanding, try posting a YouTube link into AI studio ai.dev and asking it questions about the video. You will be amazed!

although the vision leaderboard doesn't capture every vision use case, 60+ elo points reflects the significant step in core vision capabilities like transcription, spatial understanding, reading charts/diagrams & many more. Still a lot more to do, but 2.5 Pro is the best vision…

Recently had the pleasure of lecturing back at Princeton in a grad seminar. I took the opportunity to cover how scaling laws have evolved since their inception, leaning heavily on great external content from my colleagues @borgeaud_s @jalayrac @jacobaustin132 . Content in thread

The Gemini team cooked hard with Gemini 2.5 Pro, it's an awesome model that continues to lead @lmarena_ai - huge congrats to the team! Try it for yourself in the @GeminiApp now. Can't wait for you all to see what else we've been cooking 👀

The pricing for 2.5 Pro is out. Here is the pareto performance of price : lmsys elo visualized as a rainbow.