Google unveils a next-gen family of AI reasoning models

On Tuesday, Google presented Gemini 2.5, a new family of AI thinking models, which caused “thinking” before answering a question.

To start the new family of the models, Google says the company’s claimed multimodal, reasoning AI model is still the smartest model. This model will be developed on the company’s developer platform, as well as Google EU Studio, as well as the Gemini application for the 20-month AI AI plan for subscribers.

Forward, Google, said that all new AI models will have reasonable opportunities.

Since Openai launched The first AI justification model in September 2024O1, the technology industry competed to adapt or exceed the model of the model. Today, anthropic, DeepSeek, Google and XAI have AI thinking models of additional calculation and time through problems before all of the problems.

Touch methods helped to reach new heights in the math and coding tasks of AI models. Many of the Tech World believe that the AI ​​agents of thinking models will be the main component of autonomous systems that can perform the tasks. However, these models are expensive.

Google has previously practiced with AI thinking models before leaving the “Think of Thought” in December. However, the Gemini 2.5 still represents the most serious efforts of the series of “O” models.

Google claims that Gemini 2.5 Pro is superior to previous border AI models and some leading competition AI models, several criteria. In particular, Google says Excel 2.5 to Excel in creating visually impaired web applications and agent coding applications.

Google, an assessment code called Aider Polyglot, Google Deani, Anthropic and Chinese AI Lab DeepSeek from the best AI models from DeepSEEK 68.6% Programs.

However, another test measurement program is confirmed by dev skills, swe-bench 2.5 PRO, Openai’s O3-Mini and Deepseek’s R1, but 70.3% of the anthropic Claude 3.7 fell under the Sonnet.

Multimodal test consisting of thousands of tradition questions related to mathematics, humanitarian and natural sciences in mankind, Google, which performs better than most rival flagship models, has 18.8% of Gemini 2.8%.

To get started, Google says that the Gemini 2.5 Pro is sent with 1 million Token Context window, which means it can get about 750,000 words on a road. This is longer than a book series “Owner of the Rings”. Soon the Gemini 2.5 Pro will double the input length (2 million Token).

Google did not publish API prices for Gemini 2.5 Pro. The company says it will share more in the coming weeks.