Google has officially launched Gemini 2.5, its latest and most advanced artificial intelligence model, promising significant improvements in reasoning and performance. This release comes as competition in the AI industry heats up, with major players like Chinese AI firm DeepSeek and OpenAI pushing the boundaries of innovation.
Google claims that Gemini 2.5 is its most intelligent AI model to date. The experimental Gemini 2.5 Pro version has reportedly achieved state-of-the-art results across multiple benchmarks. Notably, it topped the LMArena leaderboard, which measures human preferences, by a substantial margin.
Smarter AI with Advanced Reasoning
According to Google, Gemini 2.5 models are designed as “thinking models,” capable of reasoning through thoughts before generating responses. This approach enhances performance and accuracy, allowing the model to go beyond basic classification and prediction. It can now perform complex analysis, logical reasoning, and contextual understanding, enabling more informed decision-making.
Google highlighted its longstanding efforts to make AI smarter through techniques such as reinforcement learning and chain-of-thought prompting. It previously introduced Gemini 2.0 Flash Thinking, its first thinking model, which laid the foundation for this latest iteration.
“With Gemini 2.5, we’ve reached a new level of performance by combining an enhanced base model with improved post-training. Moving forward, we are building these thinking capabilities into all of our models,” Google stated.
Key Features of Gemini 2.5 Pro
Google emphasized that Gemini 2.5 Pro delivers top-tier performance across benchmarks requiring advanced reasoning. Without relying on expensive test-time techniques like majority voting, it leads in math and science benchmarks such as GPQA and AIME 2025.
The model also scored 18.8% on Humanity’s Last Exam, a dataset designed by subject matter experts to test human-level knowledge and reasoning. Additionally, Gemini 2.5 Pro excels at creating visually compelling web apps and agentic code applications. On SWE-Bench Verified, the industry standard for code evaluation, it scored 63.8% using a custom agent setup.
Gemini 2.5 Pro is now available in Google AI Studio and the Gemini app for advanced users. Google also announced its upcoming availability on Vertex AI, with pricing details for scaled production to follow soon.
Competitive Landscape
Meanwhile, DeepSeek has gained attention for its efficient reasoning models, increasing competition for Western AI firms. Its focus on open-source models has further intensified the competitive environment.
OpenAI continues to evolve its GPT models, maintaining a key rivalry with Google. The ongoing competition between these tech giants remains a central storyline in the fast-evolving AI industry.