Google develops a next family generation of reasoning models

Tuesday, Google Gemini Google 2.5, a new family of reasoning patterns to “think” before answering a question.

To calculate the new models family, Google is launched 2.5 Pro Experimental Gemini, a Multiminal Patter, Reasonable that the company also has its smarter pattern. This pattern will not be available on the development of the development development, Google studio, and as in the Gemini Apps for the Ai’s Plan of the Asi, Gemini Asi.

Get forward, google says all their new patterns have the ability to beat.

Since opened up the First Model of Reasoning AI in September 202O1, the technical industry reassessed to match or exceed the capabilities of the model with their. Today Ethropic, Anthropic, Deeve Aleven Ali Radioing Models, who used the infar’s infar deplus and reason to make a shipment.

Reasoning technicians helped the models to realize new heights in math and coding tasks. Many in the World Tech believe in reasoning models will be a key component of the Agents of AI, Autonomous systems that can do more largely competing. However, these models are even more expensive.

Google has experienced with reasoning models first, release a “Think” gemini “version. But Gemini 2.5 represents the most serious company of the company to series “or” of orai models.

Google claims that Gemini 2.5 Pro has overcome their previous to previous models, and some of the patterned patterns, on multiple benchmarks. Specifically, Google says created Gemini 2.5 to Excel to Creating visual web applications and coding agents

On an evaluation editing an evaluation of a rating, Gemini 2.5 Protes 68.6%, USCVERMinant Top The Lab DeezaEk, antenopic, and Chinese Lab Dapen.

However, on a different detort of measuring devopifet, Gemini 2.5 Pro Partitues 63.8%, expected the anthropary of anthropary

The last examination of humanity, a mildews composed of thousands of the maths of maths, the maths, the Gemini’s Uomols 18.8%, performing the most rival flags.

To start, Google says 2,5 Pro shipping with a 1 million contract, which means that model ai can take in about 750,000 words in a single time. It’s longer that all the “lord of the” series of books. And soon, gemini 2.5 pro will double doubling the donu to input (2 million tokens).

Google does not publish API prices for Gemini 2.5 Pro. The company says that shares more in the weeks to come.