Today, we’re releasing the stable version of Gemini 2.5 Flash-Lite, our fastest and lowest cost ($0.10 input per 1M, $0.40 output per 1M) model in the Gemini 2.5 model family. We built 2.5 Flash-Lite to push the frontier of intelligence per dollar, with native reasoning capabilities that can be optionally toggled on for more demanding use cases. Building on the momentum of 2.5 Pro and 2.5 Flash, this model rounds out our set of 2.5 models that are ready for scaled production use.Our most cost-efficient and fastest 2.5 model yet

Gemini 2.5 Flash-Lite strikes a balance between performance and cost, without compromising on quality, particularly for latency-sensitive tasks like translation and classification.Here’s what makes it stand out:Best in-class speed: Gemini 2.5 Flash-Lite has lower latency than both 2.0 Flash-Lite and 2.0 Flash on a broad sample of prompts.Cost-efficiency: It’s our lowest-cost 2.5 model yet, priced at $0.10 / 1M input tokens and $0.40 output tokens, allowing you to handle large volumes of requests affordably. We have also reduced audio input pricing by 40% from the preview launch.Smart and small: It demonstrates all-around higher quality than 2.0 Flash-Lite across a wide range of benchmarks, including coding, math, science, reasoning, and multimodal understanding.Fully featured: When you build with 2.5 Flash-Lite, you get access to a 1 million-token context window, controllable thinking budgets, and support for native tools like Grounding with Google Search, Code Execution, and URL Context.Gemini 2.5 Flash-Lite in actionSince the launch of 2.5 Flash-Lite, we have already seen some incredibly successful deployments, here are some of our favorites:Satlyt is building a decentralized space computing platform that will transform how satellite data is processed and utilized for real-time summarization of in-orbit telemetry, autonomous task management, and satellite-to-satellite communication parsing. 2.5 Flash-Lite’s speed has enabled a 45% reduction in latency for critical onboard diagnostics and a 30% decrease in power consumption compared to their baseline models.HeyGen uses AI to create avatars for video content and leverages Gemini 2.5 Flash-Lite to automate video planning, analyze and optimize content, and translate videos into over 180 languages. This allows them to provide global, personalized experiences for their users.DocsHound turns product demos into documentation by using Gemini 2.5 Flash-Lite to process long videos and extract thousands of screenshots with low latency. This transforms footage into comprehensive documentation and training data for AI agents much faster than traditional methods.Evertune helps brands understand how they are represented across AI models. Gemini 2.5 Flash-Lite is a game-changer for them, dramatically speeding up analysis and report generation. Its fast performance allows them to quickly scan and synthesize large volumes of model output to provide clients with dynamic, timely insights.You can start using 2.5 Flash-Lite by specifying “gemini-2.5-flash-lite” in your code. If you are using the preview version, you can switch to “gemini-2.5-flash-lite” which is the same underlying model. We plan to remove the preview alias of Flash-Lite on August 25th.Ready to start building? Try the stable version of Gemini 2.5 Flash-Lite now in Google AI Studio and Vertex AI.

The first Gemma model launched early last year and has since grown into a thriving Gemmaverse of over 160 million collective downloads. This ecosystem includes our family of over a dozen specialized models for everything from safeguarding to medical applications and, most inspiringly, the countless innovations from the community. From innovators like Roboflow building enterprise computer vision to the Institute of Science Tokyo creating highly-capable Japanese Gemma variants, your work has shown us the path forward.Building on this incredible momentum, we’re excited to announce the full release of Gemma 3n. While last month’s preview offered a glimpse, today unlocks the full power of this mobile-first architecture. Gemma 3n is designed for the developer community that helped shape Gemma. It’s supported by your favorite tools including Hugging Face Transformers, llama.cpp, Google AI Edge, Ollama, MLX, and many others, enabling you to fine-tune and deploy for your specific on-device applications with ease. This post is the developer deep dive: we’ll explore some of the innovations behind Gemma 3n, share new benchmark results, and show you how to start building today.What’s new in Gemma 3n?Gemma 3n represents a major advancement for on-device AI, bringing powerful multimodal capabilities to edge devices with performance previously only seen in last year’s cloud-based frontier models.

Achieving this leap in on-device performance required rethinking the model from the ground up. The foundation is Gemma 3n’s unique mobile-first architecture, and it all starts with MatFormer.MatFormer: One model, many sizesAt the core of Gemma 3n is the MatFormer (🪆Matryoshka Transformer) architecture, a novel nested transformer built for elastic inference. Think of it like Matryoshka dolls: a larger model contains smaller, fully functional versions of itself. This approach extends the concept of Matryoshka Representation Learning from just embeddings to all transformer components.

During the MatFormer training of the 4B effective parameter (E4B) model, a 2B effective parameter (E2B) sub-model is simultaneously optimized within it, as shown in the figure above. This provides developers two powerful capabilities and use cases today:1: Pre-extracted models: You can directly download and use either the main E4B model for the highest capabilities, or the standalone E2B sub-model which we have already extracted for you, offering up to 2x faster inference.2: Custom sizes with Mix-n-Match: For more granular control tailored to specific hardware constraints, you can create a spectrum of custom-sized models between E2B and E4B using a method we call Mix-n-Match. This technique allows you to precisely slice the E4B model’s parameters, primarily by adjusting the feed forward network hidden dimension per layer (from 8192 to 16384) and selectively skipping some layers. We are releasing the MatFormer Lab, a tool that shows how to retrieve these optimal models, which were identified by evaluating various settings on benchmarks like MMLU.

MMLU scores for the pre-trained Gemma 3n checkpoints at different model sizes (using Mix-n-Match)

Looking ahead, the MatFormer architecture also paves the way for elastic execution. While not part of today’s launched implementations, this capability allows a single deployed E4B model to dynamically switch between E4B and E2B inference paths on the fly, enabling real-time optimization of performance and memory usage based on the current task and device load.Per-Layer Embeddings (PLE): Unlocking more memory efficiencyGemma 3n models incorporate Per-Layer Embeddings (PLE). This innovation is tailored for on-device deployment as it dramatically improves model quality without increasing the high-speed memory footprint required on your device’s accelerator (GPU/TPU).While the Gemma 3n E2B and E4B models have a total parameter count of 5B and 8B respectively, PLE allows a significant portion of these parameters (the embeddings associated with each layer) to be loaded and computed efficiently on the CPU. This means only the core transformer weights (approximately 2B for E2B and 4B for E4B) need to sit in the typically more constrained accelerator memory (VRAM).

With Per-Layer Embeddings, you can use Gemma 3n E2B while only having ~2B parameters loaded in your accelerator.

KV Cache sharing: Faster long-context processingProcessing long inputs, such as the sequences derived from audio and video streams, is essential for many advanced on-device multimodal applications. Gemma 3n introduces KV Cache Sharing, a feature designed to significantly accelerate time-to-first-token for streaming response applications.KV Cache Sharing optimizes how the model handles the initial input processing stage (often called the “prefill” phase). The keys and values of the middle layer from local and global attention are directly shared with all the top layers, delivering a notable 2x improvement on prefill performance compared to Gemma 3 4B. This means the model can ingest and understand lengthy prompt sequences much faster than before.Audio understanding: Introducing speech to text and translationGemma 3n uses an advanced audio encoder based on the Universal Speech Model (USM). The encoder generates a token for every 160ms of audio (about 6 tokens per second), which are then integrated as input to the language model, providing a granular representation of the sound context.This integrated audio capability unlocks key features for on-device development, including:Automatic Speech Recognition (ASR): Enable high-quality speech-to-text transcription directly on the device.Automatic Speech Translation (AST): Translate spoken language into text in another language.We’ve observed particularly strong AST results for translation between English and Spanish, French, Italian, and Portuguese, offering great potential for developers targeting applications in these languages. For tasks like speech translation, leveraging Chain-of-Thought prompting can significantly enhance results. Here’s an example:

user
Transcribe the following speech segment in Spanish, then translate it into English:

model

Plain text

At launch time, the Gemma 3n encoder is implemented to process audio clips up to 30 seconds. However, this is not a fundamental limitation. The underlying audio encoder is a streaming encoder, capable of processing arbitrarily long audios with additional long form audio training. Follow-up implementations will unlock low-latency, long streaming applications.MobileNet-V5: New state-of-the-art vision encoderAlongside its integrated audio capabilities, Gemma 3n features a new, highly efficient vision encoder, MobileNet-V5-300M, delivering state-of-the-art performance for multimodal tasks on edge devices.Designed for flexibility and power on constrained hardware, MobileNet-V5 gives developers:Multiple input resolutions: Natively supports resolutions of 256×256, 512×512, and 768×768 pixels, allowing you to balance performance and detail for your specific applications.Broad visual understanding: Co-trained on extensive multimodal datasets, it excels at a wide range of image and video comprehension tasks.High throughput: Processes up to 60 frames per second on a Google Pixel, enabling real-time, on-device video analysis and interactive experiences.This level of performance is achieved with multiple architectural innovations, including:An advanced foundation of MobileNet-V4 blocks (including Universal Inverted Bottlenecks and Mobile MQA).A significantly scaled up architecture, featuring a hybrid, deep pyramid model that is 10x larger than the biggest MobileNet-V4 variant.A novel Multi-Scale Fusion VLM adapter that enhances the quality of tokens for better accuracy and efficiency.Benefiting from novel architectural designs and advanced distillation techniques, MobileNet-V5-300M substantially outperforms the baseline SoViT in Gemma 3 (trained with SigLip, no distillation). On a Google Pixel Edge TPU, it delivers a 13x speedup with quantization (6.5x without), requires 46% fewer parameters, and has a 4x smaller memory footprint, all while providing significantly higher accuracy on vision-language tasksWe’re excited to share more about the work behind this model. Look out for our upcoming MobileNet-V5 technical report, which will deep dive into the model architecture, data scaling strategies, and advanced distillation techniques.Making Gemma 3n accessible from day one has been a priority. We’re proud to partner with many incredible open source developers to ensure broad support across popular tools and platforms, including contributions from teams behind AMD, Axolotl, Docker, Hugging Face, llama.cpp, LMStudio, MLX, NVIDIA, Ollama, RedHat, SGLang, Unsloth, and vLLM.But this ecosystem is just the beginning. The true power of this technology is in what you will build with it. That’s why we’re launching the Gemma 3n Impact Challenge. Your mission: use Gemma 3n’s unique on-device, offline, and multimodal capabilities to build a product for a better world. With $150,000 in prizes, we’re looking for a compelling video story and a “wow” factor demo that shows real-world impact. Join the challenge and help build a better future.Get started with Gemma 3n todayReady to explore the potential of Gemma 3n today? Here’s how:Experiment directly: Use Google AI Studio to try Gemma 3n in just a couple of clicks. Gemma models can also be deployed directly to Cloud Run from AI Studio.Learn & integrate: Dive into our comprehensive documentation to quickly integrate Gemma into your projects or start with our inference and fine-tuning guides.

Healthcare is increasingly embracing AI to improve workflow management, patient communication, and diagnostic and treatment support. It’s critical that these AI-based systems are not only high-performing, but also efficient and privacy-preserving. It’s with these considerations in mind that we built and recently released Health AI Developer Foundations (HAI-DEF). HAI-DEF is a collection of lightweight open models designed to offer developers robust starting points for their own health research and application development. Because HAI-DEF models are open, developers retain full control over privacy, infrastructure and modifications to the models. In May of this year, we expanded the HAI-DEF collection with MedGemma, a collection of generative models based on Gemma 3 that are designed to accelerate healthcare and lifesciences AI development.Today, we’re proud to announce two new models in this collection. The first is MedGemma 27B Multimodal, which complements the previously-released 4B Multimodal and 27B text-only models by adding support for complex multimodal and longitudinal electronic health record interpretation. The second new model is MedSigLIP, a lightweight image and text encoder for classification, search, and related tasks. MedSigLIP is based on the same image encoder that powers the 4B and 27B MedGemma models.MedGemma and MedSigLIP are strong starting points for medical research and product development. MedGemma is useful for medical text or imaging tasks that require generating free text, like report generation or visual question answering. MedSigLIP is recommended for imaging tasks that involve structured outputs like classification or retrieval. All of the above models can be run on a single GPU, and MedGemma 4B and MedSigLIP can even be adapted to run on mobile hardware.Full details of MedGemma and MedSigLIP development and evaluation can be found in the MedGemma technical report.

AcknowledgementsThis work was developed by the Gemini Robotics team: Abbas Abdolmaleki, Saminda Abeyruwan, Joshua Ainslie, Jean-Baptiste Alayrac, Montserrat Gonzalez Arenas, Ashwin Balakrishna, Nathan Batchelor, Alex Bewley, Jeff Bingham, Michael Bloesch, Konstantinos Bousmalis, Philemon Brakel, Anthony Brohan, Thomas Buschmann, Arunkumar Byravan, Serkan Cabi, Ken Caluwaerts, Federico Casarini, Christine Chan, Oscar Chang, London Chappellet-Volpini, Jose Enrique Chen, Xi Chen, Hao-Tien Lewis Chiang, Krzysztof Choromanski, Adrian Collister, David B. D’Ambrosio, Sudeep Dasari, Todor Davchev, Meet Kirankumar Dave, Coline Devin, Norman Di Palo, Tianli Ding, Carl Doersch, Adil Dostmohamed, Yilun Du, Debidatta Dwibedi, Sathish Thoppay Egambaram, Michael Elabd, Tom Erez, Xiaolin Fang, Claudio Fantacci, Cody Fong, Erik Frey, Chuyuan Fu, Ruiqi Gao, Marissa Giustina, Keerthana Gopalakrishnan, Laura Graesser, Oliver Groth, Agrim Gupta, Roland Hafner, Steven Hansen, Leonard Hasenclever, Sam Haves, Nicolas Heess, Brandon Hernaez, Alex Hofer, Jasmine Hsu, Lu Huang, Sandy H. Huang, Atil Iscen, Mithun George Jacob, Deepali Jain, Sally Jesmonth, Abhishek Jindal, Ryan Julian, Dmitry Kalashnikov, Stefani Karp, Matija Kecman, J. Chase Kew, Donnie Kim, Frank Kim, Junkyung Kim, Thomas Kipf, Sean Kirmani, Ksenia Konyushkova, Yuheng Kuang, Thomas Lampe, Antoine Laurens, Tuan Anh Le, Isabel Leal, Alex X. Lee, Tsang-Wei Edward Lee, Guy Lever, Jacky Liang, Li-Heng Lin, Fangchen Liu, Shangbang Long, Caden Lu, Sharath Maddineni, Anirudha Majumdar, Kevis-Kokitsi Maninis, Andrew Marmon, Sergio Martinez, Assaf Hurwitz Michaely, Niko Milonopoulos, Joss Moore, Robert Moreno, Michael Neunert, Francesco Nori, Joy Ortiz, Kenneth Oslund, Carolina Parada, Emilio Parisotto, Peter Pastor Sampedro, Acorn Pooley, Thomas Power, Alessio Quaglino, Haroon Qureshi, Rajkumar Vasudeva Raju, Helen Ran, Dushyant Rao, Kanishka Rao, Isaac Reid, David Rendleman, Krista Reymann, Miguel Rivas, Francesco Romano, Yulia Rubanova, Pannag R Sanketi, Dhruv Shah, Mohit Sharma, Kathryn Shea, Mohit Shridhar, Charles Shu, Vikas Sindhwani, Sumeet Singh, Radu Soricut, Rachel Sterneck, Ian Storz, Razvan Surdulescu, Jie Tan, Jonathan Tompson, Saran Tunyasuvunakool, Jake Varley, Grace Vesom, Giulia Vezzani, Maria Bauza Villalonga, Oriol Vinyals, René Wagner, Ayzaan Wahid, Stefan Welker, Paul Wohlhart, Chengda Wu, Markus Wulfmeier, Fei Xia, Ted Xiao, Annie Xie, Jinyu Xie, Peng Xu, Sichun Xu, Ying Xu, Zhuo Xu, Jimmy Yan, Sherry Yang, Skye Yang, Yuxiang Yang, Hiu Hong Yu, Wenhao Yu, Li Yang Ku, Wentao Yuan, Yuan Yuan, Jingwei Zhang, Tingnan Zhang, Zhiyuan Zhang, Allan Zhou, Guangyao Zhou and Yuxiang Zhou.We’d also like to thank: Amy Nommeots-Nomm, Ashley Gibb, Bhavya Sukhija, Bryan Gale, Catarina Barros, Christy Koh, Clara Barbu, Demetra Brady, Hiroki Furuta, Jennie Lees, Kendra Byrne, Keran Rong, Kevin Murphy, Kieran Connell, Kuang-Huei Lee, M. Emre Karagozler, Martina Zambelli, Matthew Jackson, Michael Noseworthy, Miguel Lázaro-Gredilla, Mili Sanwalka, Mimi Jasarevic, Nimrod Gileadi, Rebeca Santamaria-Fernandez, Rui Yao, Siobhan Mcloughlin, Sophie Bridgers, Stefano Saliceti, Steven Bohez, Svetlana Grant, Tim Hertweck, Verena Rieser, Yandong Ji.For their leadership and support of this effort, we’d like to thank: Jean-Baptiste Alayrac, Zoubin Ghahramani, Koray Kavukcuoglu and Demis Hassabis. We’d like to recognize the many teams across Google and Google DeepMind that have contributed to this effort including Legal, Marketing, Communications, Responsibility and Safety Council, Responsible Development and Innovation, Policy, Strategy and Operations, and our Business and Corporate Development teams. We’d like to thank everyone on the Robotics team not explicitly mentioned above for their continued support and guidance. Finally, we’d like to thank the Apptronik team for their support.

Introducing the first model for contextualizing ancient inscriptions, designed to help historians better interpret, attribute and restore fragmentary texts.Writing was everywhere in the Roman world — etched onto everything from imperial monuments to everyday objects. From political graffiti, love poems and epitaphs to business transactions, birthday invitations and magical spells, inscriptions offer modern historians rich insights into the diversity of everyday life across the Roman world.Often, these texts are fragmentary, weathered or deliberately defaced. Restoring, dating and placing them is nearly impossible without contextual information, especially when comparing similar inscriptions.Today, we’re publishing a paper in Nature introducing Aeneas, the first artificial intelligence (AI) model for contextualizing ancient inscriptions.When working with ancient inscriptions, historians traditionally rely on their expertise and specialized resources to identify “parallels” — which are texts that share similarities in wording, syntax, standardized formulas or provenance.Aeneas greatly accelerates this complex and time-consuming work. It reasons across thousands of Latin inscriptions, retrieving textual and contextual parallels in seconds that allow historians to interpret and build upon the model’s findings.

AcknowledgmentsGenie 3 was made possible due to key research and engineering contributions from Phil Ball, Jakob Bauer, Frank Belletti, Bethanie Brownfield, Ariel Ephrat, Shlomi Fruchter, Agrim Gupta, Kristian Holsheimer, Aleks Holynski, Jiri Hron, Christos Kaplanis, Marjorie Limont, Matt McGill, Yanko Oliveira, Jack Parker-Holder, Frank Perbet, Guy Scully, Jeremy Shar, Stephen Spencer, Omer Tov, Ruben Villegas, Emma Wang and Jessica Yung.We thank Andrew Audibert, Cip Baetu, Jordi Berbel, David Bridson, Jake Bruce, Gavin Buttimore, Sarah Chakera, Bilva Chandra, Paul Collins, Alex Cullum, Bogdan Damoc, Vibha Dasagi, Maxime Gazeau, Charles Gbadamosi, Shan Han, Woohyun Han, Ed Hirst, Ashyana Kachra, Lucie Kerley, Kristian Kjems, Eva Knoepfel, Vika Koriakin, Jessica Lo, Cong Lu, Zeb Mehring, Alexandre Moufarek, Henna Nandwani, Valeria Oliveira, Fabio Pardo, Jane Park, Andrew Pierson, Ben Poole, Helen Ran, Nilesh Ray, Tim Salimans, Manuel Sanchez, Igor Saprykin, Amy Shen, Sailesh Sidhwani, Duncan Smith, Joe Stanton, Hamish Tomlinson, Dimple Vijaykumar, Luyu Wang, Piers Wingfield, Nat Wong, Keyang Xu, Christopher Yew, Nick Young and Vadim Zubov for their invaluable partnership in developing and refining key components of this project.Thanks to Tim Rocktäschel, Satinder Singh, Adrian Bolton, Inbar Mosseri, Aäron van den Oord, Douglas Eck, Dumitru Erhan, Raia Hadsell, Zoubin Gharamani, Koray Kavukcuoglu and Demis Hassabis for their insightful guidance and support throughout the research process.Feature video was produced by Suz Chambers, Matthew Carey, Alex Chen, Andrew Rhee, JR Schmidt, Scotch Johnson, Heysu Oh, Kaloyan Kolev, Arden Schager, Sam Lawton, Hana Tanimura, Zach Velasco, Ben Wiley, and Dev Valladares. Including samples generated by Signe Nørly, Eleni Shaw, Andeep Toor, Gregory Shaw, and Irina Blok.We thank Frederic Besse, Tim Harley and the rest of the SIMA team for access to a recent version of their agent.Finally, we extend our gratitude to Mohammad Babaeizadeh, Gabe Barth-Maron, Parker Beak, Jenny Brennan, Tim Brooks, Max Cant, Harris Chan, Jeff Clune, Kaspar Daugaard, Dumitru Erhan, Ashley Feden, Simon Green, Nik Hemmings, Michael Huber, Jony Hudson, Dirichi Ike-Njoku, Hernan Moraldo, Bonnie Li, Simon Osindero, Georg Ostrovski, Ryan Poplin, Alex Rizkowsky, Giles Ruscoe, Ana Salazar, Guy Simmons, Jeff Stanway, Metin Toksoz-Exley, Xinchen Yan, Petko Yotov, Mingda Zhang and Martin Zlocha for their insights and support.