On Tuesday, Google announced Gemini 2.5, a new AI reasoning model that stopped “thinking” before answering questions.
To launch a new family of models, Google is launching the Gemini 2.5 Pro experiment, a multimodal AI model the company claims to be its smartest model to date. The model will be available on Tuesday in the company’s developer platform, Google AI Studio, as well as the Gemini App, for the company’s $20 priced AI plan, Gemini Advance.
Google expands, and Google says all its new AI models will have baked inference capabilities.
Since the launch of Openai The first AI inference model in September 2024O1, the technology industry has competed to match or exceed the capabilities of this model. Today, anthropomorphism, DeepSeek, Google, and XAI have AI inference models that utilize additional computing power and time to examine facts and reasoning through questions before providing answers.
Inference technology helps AI models reach new heights in mathematical and coding tasks. Many in the tech world believe that inference models will be a key component of AI agents, i.e. autonomous systems that can perform SAN human intervention tasks to a large extent. However, these models are also more expensive.
Google has tried AI inference models before, and previously released the “thinking” version of Gemini in December. But the Gemini 2.5 represents the company’s most serious attempt at Besting Openai’s O-series model.
Google claims that the Gemini 2.5 Pro outperforms its previous boundary AI models in several benchmarks as well as some leading AI models. Specifically, Google says it designed Gemini 2.5 to excel in creating visually engaging web applications and proxy coding applications.
Google said when evaluating measurement code editing, called Aider Polyglot, that the Gemini 2.5 Pro scored 68.6%, outperforming the top AI models of OpenAI, humans and China’s AI Lab DeepSeek.
However, on another measurement and measurement software, DEV functionality, SWE-Bench is verified, and the Gemini 2.5 Pro scores 63.8%, outperforming OpenAI’s O3-Mini and DeepSeek’s R1, but underperforming Anthropic’s Claude 3.7 sonnet, with a score of 70.3%.
In the final exam for humans, a multimodal exam consisting of thousands of crowdsourcing questions related to mathematics, humanities and natural sciences, Google says the Gemini 2.5 Pro scored 18.8%, outperforming most competitors’ flagship models.
First, Google says the Gemini 2.5 Pro ships with a 1 million token context window, meaning that the AI model can take up about 750,000 words in a single go. This is longer than the entire Lord of the Rings book series. Soon, Gemini 2.5 Pro will support twice the input length (20 million tokens).
Google has not released API pricing for Gemini 2.5 Pro. The company said it will share more in the coming weeks.
