Google Introduces Its Most Advanced AI Model Yet

Google has launched Gemini 2.5 Pro Experimental, a new artificial intelligence model designed to reason through problems before delivering answers, a shift that marks a major leap in AI capability, according to the company.

The model, available immediately through Google AI Studio and the Gemini app for subscribers to the company's $20-per-month Gemini Advanced plan, is part of the new Gemini 2.5 family of "thinking" models. It's built to analyze information, weigh context, and make informed decisions across a range of complex tasks including code generation, scientific reasoning, and multimodal interpretation of images, audio, and video.

Google said Gemini 2.5 Pro outperforms several rival models, including those from OpenAI and DeepSeek, in key benchmarks. On a code editing test called Aider Polyglot, Gemini 2.5 Pro scored 68.6%. On SWE-bench Verified, a measure of software development skills, it scored 63.8%, surpassing some competitors but trailing Anthropic's Claude 3.7 Sonnet, which scored 70.3%.

The company said the model also achieved 18.8% on Humanity's Last Exam, a rigorous multimodal test of general knowledge and reasoning. Gemini 2.5 Pro currently supports a 1 million token context window — enough to process roughly 750,000 words at once — and plans to expand to 2 million tokens in the near future.

Google did not immediately announce pricing for API access to the new model but said more details would be shared in the coming weeks.

The launch comes amid growing competition in the AI sector, particularly around reasoning models. Since OpenAI introduced its "o" series in late 2024, companies including Anthropic, xAI, and DeepSeek have released models with similar capabilities. Unlike earlier systems focused mainly on pattern recognition or prediction, reasoning models aim to methodically process information — a feature seen as foundational for autonomous AI agents.

Gemini 2.5 Pro is Google's most serious effort yet to surpass competitors in the rapidly advancing AI landscape. The company said reasoning capabilities will be included in all future Gemini models.

For more information, visit the Google blog.

About the Author

John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS.  He can be reached at [email protected].

Featured

  • DreamBox Math

    Discovery Education Announces Updates to Experience, DreamBox Math

    K-12 learning solution provider Discovery Education has announced enhancements to its Discovery Education Experience and DreamBox Math products, designed to create a more personalized, engaging learning experience for students.

  • abstract pattern of cybersecurity, ai and cloud imagery

    Report Identifies Malicious Use of AI in Cloud-Based Cyber Threats

    A recent report from OpenAI identifies the misuse of artificial intelligence in cybercrime, social engineering, and influence operations, particularly those targeting or operating through cloud infrastructure. In "Disrupting Malicious Uses of AI: June 2025," the company outlines how threat actors are weaponizing large language models for malicious ends — and how OpenAI is pushing back.

  • digital dashboard featuring a shield icon, graphs, a world map, and network nodes

    IBM Launches Agentic AI Governance and Security Platform

    IBM has introduced a new software stack for enterprise IT teams tasked with managing the complex governance and security challenges posed by autonomous AI systems.

  • laptop and fish hook

    Security Researchers Identify Generative AI 'Vishing' Attack

    A new report from researchers at Ontinue's Cyber Defense Center has identified a complex, multi-stage cyber attack that leveraged social engineering, remote access tools, and signed binaries to infiltrate and persist within a target network.