Companies

Google Recaps 2025: Gemini 3, AlphaFold Milestones and a Physics Nobel

Google's year-end review claims Gemini 3 Pro topped LMArena, AlphaFold passed 3 million researchers, and a Googler won a Physics Nobel — AI's shift from tool to utility.

Google's year in review: 8 areas with research breakthroughs in 2025
Google's year in review: 8 areas with research breakthroughs in 2025AI-generated
By Elena Vasquez3 min read

Updated

Why it matters

  • Gemini 3 Pro topped the LMArena Leaderboard, set a 23.4% state-of-the-art on MathArena Apex, and Gemini 3 Flash launched in December at a fraction of Gemini 2.5 Pro's price
  • AlphaFold, five years old in 2025, has been used by over 3 million researchers in more than 190 countries, including over 1 million in low- and middle-income countries
  • Flood forecasting now covers more than two billion people in 150 countries; WeatherNext 2 generates forecasts 8x faster with up to 1-hour resolution

Google launched Gemini 3 Pro in November, topped the LMArena Leaderboard, and set a new state-of-the-art of 23.4% on MathArena Apex — the headline numbers in a year-end review the company published covering Google, Google DeepMind and Google Research.

The recap covers eight areas, but the throughline is Google's own framing: "If 2024 was about laying the multimodal foundations for this era, 2025 was the year AI began to really think, act and explore the world alongside us." Google describes AI's trajectory as shifting "from a tool to a utility: from something people use to something they can put to work."

Model milestones. The year began with Gemini 2.5 in March and ended with Gemini 3 in November and Gemini 3 Flash in December. Gemini 3 Pro posted breakthrough scores on Humanity's Last Exam — which Google calls "a fiendishly hard test for AI models to see if AI can truly think and reason like humans" — and on GPQA Diamond. Gemini 3 Flash, per Google, surpasses the capabilities of the previous Gemini 2.5 Pro-scale model at a fraction of the price, continuing what the company calls its Gemini-era trend: "the next generation's Flash model is better than the previous generation's Pro model." The open Gemma family gained multimodal capabilities, a larger context window and expanded multilingual support.

Agents and products. Google points to Gemini 3's coding capabilities and the launch of Google Antigravity as marking "a new era in AI-assisted software development." Core products changed too: AI features on the Pixel 10, generative UI in AI Mode in Search, Deep Research in NotebookLM, and Labs experiments including Jules, an asynchronous coding agent, and Google Beam, a 3D video communications platform.

Generative media. Nano Banana and Nano Banana Pro broke out for native image generation and editing. Google also shipped Veo 3.1, Imagen 4 and Flow, and built tools with creative-industry input such as Music AI Sandbox.

Science. 2025 marked the fifth anniversary of AlphaFold, the Nobel-winning system that solved the 50-year-old protein folding problem. Google says more than 3 million researchers in over 190 countries have used it, including over 1 million in low- and middle-income countries. Other releases: AlphaGenome (June), DeepSomatic for identifying genetic variants in tumors (October), and an AI co-scientist (February). Gemini's Deep Think mode achieved gold medal-standard results in two international contests requiring deep abstract reasoning.

Quantum and hardware. Googler Michel Devoret shared the 2025 Physics Nobel with former Googler John Martinis and UC Berkeley's John Clarke for foundational 1980s quantum research. The Quantum Echoes algorithm showed progress toward real-world applications. Google also introduced Ironwood, a TPU "built for the age of inference" and designed with the AlphaChip method, and unveiled Project Suncatcher, a design study for space-based AI infrastructure. Robotics advances included Gemini Robotics 1.5 and Genie 3, which Google calls a new frontier for general-purpose world models.

Applied at scale. Flood forecasting now covers more than two billion people in 150 countries for severe riverine floods. WeatherNext 2 generates forecasts 8x faster with resolution up to 1 hour, and Google has supported weather agencies with experimental cyclone predictions. In education, initiatives include LearnLM and Guided Learning in Gemini, plus smarter translations in Google Translate.

Safety and collaboration. Google calls Gemini 3 its most secure model yet, saying it "has undergone the most comprehensive set of safety evaluations of any Google AI model to date." In December, Google DeepMind and other leading labs formed the Agentic AI Foundation, anchored by contributions including the Model Context Protocol (MCP). Google is also working with the US Department of Energy's 17 national laboratories, including on the Genesis mission announced in December, and with school districts such as Miami Dade County.

The stakes are straightforward: Google is positioning 2025's research output — frontier models, scientific tooling, quantum progress — as evidence it can convert breakthroughs into deployed utility faster than competitors. For 2026, the company says it will "continue to advance the frontier, safely and responsibly."

Original: blog.google

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Market editor covering media and advertising at AI In Context.

122 articles

Related articles

  1. Google Ships Gemini 3.1 Pro, More Than Doubling Reasoning Score
  2. Google Upgrades Gemini 3 Deep Think With Record Benchmark Runs
  3. Google Ships Gemini 3.5 Flash, Promises Pro Model Next Month
  4. Google Launches Gemini 3 Pro at $2/Million Input Tokens
  5. Google DeepMind chief says Gemini 4 is nearly ready for launch

« Previous articleNext article »