Google’s Gemini 2.5 Sets New AI Benchmarks with Advanced Thinking, Deep Research, and Superior Coding Abilities

Google has released Gemini 2.5, its most sophisticated AI model to date, which includes major advancements in reasoning, customisation, and multimodal capabilities. With its ability to provide extremely context-aware and tailored answers, this current model takes a step closer to Artificial General Intelligence (AGI).  

One of the most notable aspects of Gemini 2.5 is its expanded context window, which now supports 1 million tokens, with an upgrade to 2 million tokens planned soon. This enables the model to process large amounts of data from text, photos, audio, video, and coding, making it an effective tool for complicated work. Furthermore, its "Deep Research" have functions as an AI-powered research helper, gathering and analyzing information from the web to produce extensive reports in minutes.  

Along with Google's innovations, AI firm Anthropic recently introduced Claude 3.7 Sonnet, a hybrid model capable of both fast reactions and in-depth reasoning. However, Gemini 2.5 Pro, the first test release in the Gemini 2.5 series, has established a new LMArena leaderboard record in human choice ratings. The model excels in math, science, and programming, scoring an amazing 63.8% on the SWE-Bench Verified leaderboard for software engineering problems.  

At present, Gemini 2.5 Pro is accessible for developers and corporations through Google AI Studio and Gemini Advanced, with a wider rollout on Vertex AI coming soon. Pricing for large-scale applications will be released in the coming weeks.

This article is based on information from The Tech Portal