Google has released its primary AI research progress for 2025. The company positions this year as the period when AI transitioned from a tool to a utility. While 2024 focused on establishing multimodal foundations, Google explains that in 2025, AI began to think, act, and explore the world. Furthermore, the company stated that it focused on progress toward the practical application of quantum computing and the translation of research into real-world products and tools.
Model ReleasesGoogleGemini 3Gemma 3Gemini 3 FlashGemini 3 ProGoogle Antigravity
Google Releases 2025 AI Research Progress: Reporting Performance Gains for Gemini 3 and Gemma 3
This article is a translation. Read the Japanese original
Regarding models, starting with the release of Gemini 2.5 in March, Google announced Gemini 3 in November and Gemini 3 Flash in December. As the top-tier model, Gemini 3 Pro has secured the top spot on the LMArena leaderboard. It recorded high scores on benchmarks such as Humanity’s Last Exam and GPQA Diamond, and established a new record of 23.4% on MathArena Apex in the field of mathematics. Following this, Gemini 3 Flash provides Pro-level reasoning performance with Flash-like low latency and high efficiency. It is reported to achieve quality exceeding the previous generation's Pro model while being lower cost and faster.
For the open-weight Gemma family, progress was made in adding multimodal capabilities, expanding context windows, and strengthening multilingual support. Additionally, in the field of software development, Google reported that it deepened collaboration with developers through the introduction of "Google Antigravity," an agentic system that goes beyond coding assistance tools.
Source: Google's year in review: areas with research breakthroughs in 2025（HN 245pt・168コメント） (HN Search (backfill), 2025-12-24)