After DeepSeek launched its new AI model, Google stole the show by announcing Gemini 2.5 on March 25. Google calls it their “smartest model” yet. They said this is an “experimental version of Gemini 2.5 Pro.” It leads in various benchmarks and took first place on LMArena by a large margin at launch.”
Gemini 2.5 is part of Google’s Gemini AI family. It enhances reasoning skills and builds on Gemini 2.0 Flash Thinking, which was introduced in December 2024. The Gemini 2.5 Pro outperformed OpenAI’s o3-mini and Anthropic’s Claude 3.7 Sonnet on Humanity’s Last Exam (HLE). This new test aims to fix “benchmark saturation.” That happens when AI models advance so quickly that current tests become too easy. Achieving an 18.8% score on HLE, Gemini 2.5 outperformed o3-mini’s 14% (which assesses only text-based inputs) and Claude 3.7 Sonnet’s 8.9%.
Gemini 2.5 is currently at the top of the Chatbot Arena rankings. It shows great results in science, math, and coding tests. It beats other models by small but clear margins. This is no surprise, given how fast AI is advancing. Google says the new Gemini 2.5 Pro shows big gains in reasoning, multimodal processing, and agent-based interactions. It works well even with little user input.
こちらもお読みください： セールスフォース、Tableau PulseとTableau Agentの日本語サポートを開始
Gemini 2.5 Pro now offers a context window of 1 million tokens. As of March 25, it is available to Gemini Advanced users through Google AI Studio and Gemini apps. Soon, it will also be available on Vertex AI. グーグル has also indicated that it will announce pricing details in the coming weeks.
