Google's new AI can now “think deep”: More than just answers - faster, smarter, smarter and sounds more human than ever - Image: Xpert.digital
Gemini 2.5 makes Google the AI market leader: What this means for users
Google introduces new Gemini-2.5 model and expands access
Google has announced significant progress in its Gemini-2.5 model and significantly expanded access to its most advanced AI technology. The latest developments mark an important milestone in Google's AI strategy and position the company as a leading provider in the field of artificial intelligence.
Suitable for:
General availability of Gemini 2.5 Flash and Pro
On June 17, 2025, Google released the stable version of Gemini 2.5 Flash and Pro for general use. These models are no longer in the test phase and can be used confidently in productive applications. Developers and companies can use the models via Google AI Studio and Vertex AI, although they are also available in the Gemini app.
The general availability means that Google will support these models in the long term and that they are suitable for scaled production applications. Companies such as Spline, Rooms, Snap and Smartbear have successfully worked with the latest versions in the past few weeks.
Introduction of Gemini 2.5 Flash-Lite
As the latest addition to the Gemini-2.5 family, Google Gemini 2.5 Flash-Lite has presented the cost-effective and fastest model in the 2.5 series. Flash-Lite is initially available as a preview and is aimed at developers who have to carry out high-volume tasks such as translations and classifications.
The new model offers significantly better quality than Gemini 2.0 Flash-Lite in the areas of programming, mathematics, natural sciences, logical thinking and multimodal benchmarks. It only causes a fraction of the costs of the full price models and offers lower latency times than its predecessors.
Extended thinking skills and Deep Think
A key feature of the Gemini-2.5 models are their advanced “thinking” skills. These thinking models are able to think through their thoughts before the answer, which leads to improved performance and greater accuracy. With “Thinking Budgets”, developers can check how intensively the model should think before it generates an answer.
Google has also announced an experimental “Deep Think” mode for Gemini 2.5 Pro. This mode enables the model to pursue several ways of thinking in parallel before there is an answer, which is particularly advantageous for complex mathematical and programming tasks. In tests, Deep Think achieved top performance in the US math Olympics 2025 and achieved 84% in the demanding MMMU benchmark.
New functions and improvements
Native audio output and live api
Gemini 2.5 receives native audio output functions that enable more natural conversations. The extended live api supports audiovisual entries and enables direct conversations with the AI. Users can control the tone, accent and the expression and, for example, say the model that it should read stories in a dramatic tone.
The new experimental functions include:
- Affective Dialogue: The model recognizes emotions in the voice and answers accordingly
- Proactive audio: automatic filtering of background talks
- Text-to-Speech: Support for several speakers in over 24 languages
Improved programming skills
Gemini 2.5 Pro leads the webdev arena ranking and shows significant improvements in web development. The model reaches 63.8% in the SWE-Bench Verified, the industrial standard for agent-based code evaluations. It excelles when creating visually appealing web apps and agent-based code applications as well as code transformation and processing.
Video mm benchmark shows the impressive multimodal skills: Gemini 2.5 Pro reaches 84.8% compared to 75% in Gemini 1.5 per and 71.9% in GPT-4O. This ability makes it possible to create entire applications from video content.
Extended multimodality and context processing
Gemini 2.5 builds on the strengths of the Gemini models: native multimodality and a large context window. The model starts with a 1 million token context window, whereby 2 million tokens will be available shortly. It can understand extensive data records and edit complex problems from different information sources, including text, audio, images, videos and entire code repositories.
Availability and access
For developers
- Google Ai Studio: Immediate availability for experiments
- Vertex AI: Available for companies with extended functions
- Gemini API: Complete integration with SDK support
For end users
- Gemini app: Available for Gemini Advanced users on desktop and mobile
- Google Search: Specially adapted versions of Flash-Lite and Flash
Education sector
Google extends the free access to the Google Ai Pro-Plan for students in Brazil, Indonesia, Japan and Great Britain to the final exams 2026. In addition to AI support, the package also includes 2 TB storage space and NoteboOKKLM.
Suitable for:
Technical specifications and performance
Gemini 2.5 Pro leads the Lmarena ranking at a clear distance and shows state services in various benchmarks. The model reaches 18.8% at “Humanity's Last Exam”, a dataset developed by hundreds of experts in order to grasp the human limit of knowledge and logical thinking.
The latest version of Gemini 2.5 Pro shows a 24-point ELO jump on Lmarena and a 35-point ELO jump on webdevara. It continues to lead to difficult programming benchmarks such as Aider Polyglot and shows top performance in GPQA and other demanding mathematical and scientific reviews.
Google Gemini 2.5 Flash and per transform the AI landscape with stable versions
The introduction of the stable versions of Gemini 2.5 Flash and Pro as well as the preview of Flash-Lite mark an important step in Google's AI development. With the combination of improved performance, expanded functions and wider access, Google positions itself as a leading provider in the field of artificial intelligence.
The continuous improvements and the expansion of availability show Google's commitment to make AI technology more accessible and efficient for developers, companies and end users. With the new thinking skills and improved multimodality, Gemini 2.5 sets new standards for the next generation of AI applications.
Suitable for:
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your national language!
Konrad Wolfenstein
I would be happy to serve you and my team as a personal advisor.
You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital
I'm looking forward to our joint project.