Google (GOOG, Financials) said Google DeepMind has released Gemini 2.5 Pro, its latest artificial intelligence model, which outperformed competitors on several math, science and reasoning benchmarks, including a top score on the LMArena leaderboard, the company said on March 25.
Claiming it sets industry standards in thinking and coding, Google DeepMind presented Gemini 2.5 Pro, a new artificial intelligence model meant for challenging issue solving, on March 25.Google calls the model, which topped the LMArena ranking of AI models according to human tastes, its "most intelligent" to date. Gemini 2.5 Pro also said to get top marks in math and science tests like the Graduate-Level Physics Question Answering (GPQA) and the American Invitational Mathematics Examination (AIME) 2025 datasets.Google DeepMind claims Gemini 2.5 Pro scored 18.8% on the "Humanity's Last Exam," a benchmark created by subject matter experts to evaluate high-level reasoning capacity in AI models without external tools.Gemini 2.5 Pro, the firm said, outperformed its predecessor in coding performance. Using a bespoke agent configuration, it achieved 63.8% on SWE-Bench Verified, a test for AI-driven code creation and modification. Google said the model can turn requests into full, running web apps and entire, executable video game code.Released as an experimental version, Gemini 2.5 Pro is accessible via Google AI Studio and the Gemini app for Gemini Advanced users. The firm said it would shortly disclose prices for increased use and make access to Vertex AI of Google Cloud available.With a context window of one million tokens, the model can handle massive amounts of data from text, audio, photos, video even whole code repositories. Google forecasts a future upgrade with a 2 million token context window.Building on the company's previous "thinking model" approach, Gemini 2.5 uses chain-of-thought prompting and reinforcement learning to enhance decision-making capacity. Google said the 2.5 Pro model combines these techniques with a new base model and improved post-training.The launch follows the introduction of Gemini 2.0 Flash Thinking and continues on Google's work to integrate sophisticated reasoning directly into its models to assist more complicated use cases.Developers and business users may now experiment with Gemini 2.5 Pro. The business is also encouraging consumer input to help direct next enhancements.
