Just months after launching the Gemini 2.0 Flash Thinking model, Google is rolling out another significant AI upgrade—Gemini 2.5. The first model built on this new technology, Gemini 2.5 Pro Experimental, is now available for Advanced users and Gemini AI Studio.
Google to Infuse Reasoning on Base AI Models
In a press release, Google calls Gemini 2.5 its “most intelligent” AI model to date. Like Gemini 2.0 Flash Thinking, it belongs to the thinking model category, making it the second of its kind.
What sets Gemini 2.5 apart from earlier models is its ability to derive reasoned conclusions, apply contextual intelligence, and make well-informed decisions—moving beyond the traditional “classification and prediction” approach seen in non-thinking models.
Developed as an enhanced version of Gemini 2.0 Flash Thinking, the Gemini 2.5 model benefits from an improved base and more refined post-training. Google has also announced plans to incorporate reasoning capabilities from Gemini 2.5 into its non-thinking models.
Gemini 2.5 Pro: A New Benchmark in AI Reasoning
The first product powered by Gemini 2.5 is Gemini 2.5 Pro Experimental, which is available as a model option within Google’s AI services.
According to Google, Gemini 2.5 Pro Experimental has outperformed leading reasoning models, including OpenAI’s o3-mini and Claude 3.5 Sonnet, in Humanity’s Last Exam test—a benchmark for AI reasoning and knowledge. Additionally, Gemini 2.5 Pro achieved higher scores in math and science benchmarks, surpassing competitors like OpenAI’s GPT-4.5 and X’s Grok 3 Beta.
Google also emphasizes enhancements in coding and software development with Gemini 2.5 Pro. The model is optimized for coding, editing, and building “agentic applications and web apps.” In a demonstration video, it showcased its capabilities by generating a game from a single-line prompt.
Currently, Gemini 2.5 Pro is available to Advanced subscribers under One AI Premium, as well as business and educational plans. It is also accessible via Gemini Studio, with future availability planned for Vertex AI. While usage remains limited, Google has stated that it will announce pricing for extended access soon.
The launch of Gemini 2.5 follows Google’s expansion of Gemini 2.0 Flash and Deep Research models along with Gemini Live screen sharing to a broader audience—previously exclusive to paid users. Given this trend, it’s possible that Gemini 2.5 could become widely accessible in the near future.
Have you used Gemini’s reasoning models? Share your experience in the comments!
