Google Goes Online with Gemini 2.5 Pro Models

Early this morning, Google officially launched Gemini 2.5 Pro Experimental, the "strongest reasoning model", which performed quite well in a number of tests: it ranked No. 1 in the LMSYS Arena, with a score 40 points higher than the Grok-3 and GPT-4.5; it was ranked No. 1 in all the categories (general ability, coding, math, etc.); and it excelled in a number of benchmark tests, including Hard Prompts w/ Style Control and Multi-Turn. Gemini 2.5 Pro ranked #1 in all categories (General Skills, Coding, Math, etc.), especially in Hard Prompts w/ Style Control and Multi-Turn; Gemini 2.5 Pro took the top spot for overall performance in the individual benchmarks. Gemini 2.5 Pro is the best overall performer in all benchmarks, leading in Science, Code Generation, Visual Reasoning (MMMU), and Long Text Comprehension (MRCR); and in the so-called toughest test, "Human's Last Exam," Gemini 2.5 Pro outperforms a number of large models such as OpenAI o3-mini, GPT-4.5, DeepSeek-R1, and so on.

Search