Google unveils a next-gen family of AI reasoning models

On Tuesday, Google AI unveiled a new family of reasonable models, Gemi 2.5 that breaks to “think” before answering a question.

To remove the new families of models, Google Gemi 2.5 Pro is launching a multimodal, rational AI model that the company claims that it is still the most intelligent model. This model will be available on Tuesday at the company’s developer platform, Google AI Studios, as well as the company’s $ 20-month AI plan, the Geminee app for Gemi Advance.

Going forward, Google says that all its new AI models will have rational powers baked.

Since the launch of the Open The first AI argument model in September 2024O 1, the technology industry has been able to match or cross the capacity of that model with their own. Today, ethnographic, dipsek, google and jai have AI rational models, which use extra computing energy and time for fact-checks and logic through problems before answering.

Arguing strategies have helped AI models gain new height in mathematics and coding works. Many in the Tech World believe that rational models will be a key component of AI agents, autonomous systems that can perform work that basically suns the human intervention. However, these models are more expensive.

Google has previously tested with reasonable models, published a “thoughtful” version of Jemini in December. However, Gemi 2.5 represented the company’s most serious attempt in the best best to bestate the “O” series models.

Google claims that Gemi 2.5 Pro exceeds its previous border AI models and several top contestant AI models. Specifically, Google says it has apparently designed Gemi 2.5 to skill the creative web applications and agent coding applications.

In a measure of an evaluation known as the Ider Polyglot, Google says that Gemie 2.5 Pro score 68..6%, OpenAI, ethnographic and top AI models from the Chinese AI Lab DIPSEC.

However, in another test, in the case of measuring software dave capacity, the needle-bench has been verified, Gemi 2.5 Pro score 63.8%, Openai and 3-Minit and DIPSC have surpassed R1K, but ethnographic Claud underforming the Sonnet, which scored 70.3%.

In the last test of humanity, a multimodal test composed of thousands of public questions related to mathematics, humanity and natural science, Google has said that Gemie 2.5 Pro score is 18.8%by performing better than most rival flagship models.

To get started, Google says that Gemi 2.5 Pro is shipping with 1 million token context windows, which means the AI ​​model can take about 750,000 words alone. It’s long than a series of books “Lord of the Rings”. And soon, the Gemini 2.5 Pro input length (2 million token) will support twice.

Google Gemi has not revealed the API price for 2.5 Pro. The company says it will share more in the coming weeks.