In the latest announcements at @Google I/O Connect 2025, @osanseviero , Staff AI Developer Relations Engineer at @GoogleDeepMind , revealed a massive milestone for the Gemma family of open models over 180 million downloads and 80,000+ developer-built variants. Built on the same tech that powers Gemini, Gemma has fast become the backbone of countless AI innovations across the globe. Celebrating its first anniversary this March, Gemma's rise has been fueled by a passionate and creative developer community that continues to push the boundaries of what's possible. From long-context handling to multilingual support, the model family has become a go-to for developers eager to build smart, efficient, and context-aware applications across domains. Omar also announced the launch of Gemma 3, now the most multilingual open model on the planet, with support for over 140 languages including many Indian languages. From secure deployments using Shield Gemma to healthcare breakthroughs with Med Gemma, these tailored variants are reshaping industries and enabling safer, more powerful AI at scale. And the innovation doesn’t stop just two weeks ago, Google released a 27-billion parameter multimodal Gemma model for advanced use cases. Whether you’re building on your own hardware or creating apps for healthcare, Gemma is proving to be the open model foundation of choice for a rapidly growing global ecosystem.

Using Adaptive Engine, @SKtelecom tuned open models as small as Gemma 3 4B to exceed frontier performance (GPT-4.1, 3.7 Sonnet, and o4-mini) at multilingual content moderation. Our research 📃 and full results 👇

Build powerful resume agents with our new @crewAIInc and Gemini 2.5 quickstart, using built-in grounding with Google Search to power your Crew with a single API key. In the example: 🔎 Research your public GitHub profile 🔬 Deep dive into your projects 📝 Write a custom CV

Check out this cool MedGemma demo!

Today we are rolling out our first Gemini Embedding model, which ranks #1 on the MTEB leaderboard, as a generally available stable model. It is priced at $0.15 per million tokens and ready for at scale production use!

Excited to introduce GenAI Processors! An Open-Source Python library from @GoogleDeepMind that allows you to build asynchronous and composable AI Pipelines for Generative AI

🏥Introducing MedGemma, part 2, including: 🔥A 27B multimodal MedGemma 👀MedSigLIP, a lightweight image/text encoder for medical image retrieval/classification 📜A technical report with details Blog: Paper:

Introducing T5Gemma: the next generation of encoder-decoder/T5 models! 🔧Decoder models adapted to be encoder-decoder 🔥32 models with different combinations 🤗Available in Hugging Face and Kaggle

Introducing MatFormer Lab for Gemma 3n 🧑‍🔬 Use Mix-n-Match to slice the E4B and create a model with a custom size between 2B and 4B effective parameters Explore the quality-size trade-off and share your models with the community Try it out:

MLX-VLM v0.3.0 is here! And it brings a lot of significant improvements 🔥📷 What’s new: - KV Cache quantization - Mixed quantization (i.e, 4bit + 6bit) - Add support for audio modality in server - Gemma3n: Fixed embeddings, Vision, pixel casting, multi-task (audio + vision) and OCR - Fixed chinese decoding - Refactored generate and convert package Thank you very much and shout out to the amazing mlx-vlm contributors Neil from LMStudio, @pcuenq and Nicekate > pip install -U mlx-vlm Please leave us a start ⭐️