At the Research@ event in Mountain View, Google Research unveiled ten recent research-driven tools and projects intended to place new capabilities into the hands of scientists, clinicians, developers, public health officials and enterprise teams. The presentations emphasized AI as an amplifier of human curiosity and rigorous inquiry.
1. Google Earth AI: planetary intelligence
Google Earth AI aims to provide accurate, large-scale real-world understanding to help forecast public health threats, accelerate clean energy investment, and support business foresight. Google Research is expanding early access and offering three new ways to use experimental Earth AI capabilities: deeper geospatial analysis within an early access feature in Google Earth; risk assessments and scenario forecasting via a planetary prediction engine; and integration of Geospatial Reasoning agent skills through an experimental agent-to-agent API that gives programmatic access to satellite image parsing, routing and demographic embeddings.
The platform is powered by more than 70 geospatial tools, including foundation models such as Remote Sensing Foundation Model (RSFM) and Population Dynamics Foundation Model (PDFM). Empirical benchmarks are documented in the Earth AI Technical Report. Google Dataset Search was integrated for queries against more than 100 million public datasets, and TimesFM is used for planetary nowcasting and forecasting. To support social-impact work, Google.org is providing $10 million in funding for organizations focused on crisis resilience and public health, including $2 million to The Milken Institute to address infrastructure bottlenecks for affordable housing, clean energy, broadband, safe water systems and other essential infrastructure.
2. Multi-agent orchestration: leveraging existing literature
Google Research developed a multi-agent orchestrator to help scientists reproduce past methods and analyses and apply them in new contexts. One example used an inverse design methodology from fluid mechanics to redesign urinary catheters; simulations found catheter designs that better prevented bacterial infections. This work complements tools like Empirical Research Assistance — which can generate and test thousands of code variants in parallel — and Co-Scientist, co-developed with Google DeepMind, which can propose novel hypotheses and has been used with partners such as researchers at Stanford University to identify repurposed medicines for liver fibrosis.
3. Deep Review: improving clarity and accuracy of technical documents
Deep Review is an agentic system designed for rigorous verification of complex documents, checking logical consistency and grounding facts while leaving creative direction to authors. An early version has been used by Google researchers and external authors to review over 16,000 research submissions to top-tier computer science conferences. Access to Deep Review is being expanded to the public to help verify a variety of technical documents, from legal agreements to financial reports.
4. Next-gen Edge AI chip for wearables and edge devices
To support wearables and other edge applications, Google and Synaptics previously released a Coralboard using Google’s Coral NPU accelerator core. Next week they plan to reveal a co-designed, next-generation low-power Edge AI chip built by Synaptics using Google’s open source IP. The chip is intended to deliver improved efficiency and performance for engineers building next-generation smart glasses, watches, connected cameras, streamers, and industrial vision and voice systems.
5. TimesFM and TabFM: foundational models for numerical reasoning
TimesFM and TabFM are foundation models designed specifically to reason over raw numerical distributions: time-series data and structured tabular data. Both models provide accurate out-of-the-box predictions with zero training and rank at the top of leading public benchmarks. They are available to Cloud BigQuery users and will be integrated into Gemini for business. That integration will let business owners, analysts and operations leads forecast trends — for example product sales or hospital bed capacity — from complex operational tables containing millions of rows, while sensitive data remains within an organization’s perimeter.
6. MedGemma: open foundation for health AI applications
MedGemma, published in Nature Medicine last week, has been downloaded more than 10 million times by over 65,000 unique innovators. It serves as an open foundation for developers building specialized health applications, including faster triage in overcrowded clinics and screening for eye disease and cervical cancer. Google Research is advancing MedGemma with tool use so it can reference available health records and other clinician tools, and is exploring new modalities such as mammography, ultrasound and electrocardiograms. Early results show substantial capability gains in answering clinical questions and interpreting radiology scans, skin images and pathology slides.
7. AMIE: conversational AI to support patients and clinicians
AMIE research explores how conversational AI can assist clinical reasoning and patient–clinician conversations. Work published in Nature expanded AMIE’s capabilities beyond initial diagnosis toward long-term disease management, and Nature Medicine showed multimodal reasoning across medical images and lab results. A new study in The Lancet reports on real-world use: at Beth Israel Deaconess Medical Center, 98 patients consulted AMIE safely ahead of urgent care visits, and clinicians reported that AI summaries helped them prepare in 75% of cases. Google is also advancing AMIE’s multimodal abilities toward real-time video consultations and has launched a nationwide randomized study of AI in real-world virtual care settings in partnership with Included Health.
8. Groundsource: improving disaster prediction
Google Research is working toward a crisis-resilience goal where communities are not surprised by natural disasters. Their global hydrologic model, published in Nature, provides reliable river forecasts up to 7 days in advance for the most significant riverine floods, covering more than 2 billion people in over 150 countries. To address flash floods, they developed Groundsource, an AI-powered methodology that converts publicly available crisis information into high-quality historical datasets. Using Gemini to analyze flood-related news, they constructed a structured dataset of over 2.6 million flood events, which helped train a model that can predict the likelihood of urban flash flooding up to 24 hours in advance. That dataset is included in the Google Earth AI collection and has been made openly available for further research.
9. Learning interactives: teacher-created guided simulations
A research experiment explored generative UI with pedagogical guardrails to let teachers create custom, guided STEM simulations aligned to their curricula. A sample library includes more than 30 AI-generated learning interactives vetted by teachers for middle and high school physics, chemistry, biology and math. Schools can join the Google for Education Pilot Program to provide feedback and help improve and expand the simulations for public use.
10. Play with Putty: collaborative, real-time prototyping
Play with Putty enables teams to build and prototype together in a shared environment without disrupting creative flow. The tool demonstrates how AI models can generate full interfaces from natural language prompts, supports direct interaction with interface elements and allows multiplayer editing for real-time collaboration. Originally designed to help Google researchers and teams prototype together — for example when exploring new Google Earth AI visualizations — the waitlist for Play with Putty is open on Google Labs to broaden access.
Conclusion
These ten updates reflect progress across multiple domains of Google Research’s portfolio, providing new tools for earth observation, scientific workflows, healthcare, education and enterprise analytics. The presentations framed the technology as a way to extend, not replace, human expertise and ingenuity.
