Gemini 2.5 Pro of Google is better in coding, mathematics and science than your favorite AI model.

Google has unveiled Gemini 2.5 Pro, which is the first in its Gemini 2.5 family. It performs better than contestants from Openi, anthropropic and Deepsek in the major benchmarks related to multimodal region model coding, mathematics and science.

What is the logic of AI model?

The argument AI is “designed to think before speaking.” They evaluate reference, process details, and fact-centered reactions to ensure logical accuracy-although these abilities demand more computing power and high operational costs.

Openai launched the first argument model in the last September with a notable departure from the GPT series, focused on large -scale language production. Since then, prominent players in the AI ​​race have replied: Deepsek with R1, Anthropic with Cloud Sonnet 3.7And with xai Grocke 3,

Development beyond ‘flash thinking’

Google first launched its first Reasoning AI model, Mithun 2.0 Flash Thinking in December. Marketed for its agent abilities, flash thinking was recently File updated to allow upload And big signs; However, with the introduction of Gemini 2.5 Pro, Google is fully retiring the “thinking” label.

As Google announced about Gemini 2.5This is because logic abilities will now be originally integrated into all future models. This change marks a step towards a more integrated AI architecture rather than distinguishing “thinking” features as standalone branding.

The new experimental model “an importantly enhanced base model” combines with “better post-training”. Google LMARENA averts its performance at the top of the leaderboard, which ranks the major big language models in various tasks.

Download: How to use AI in business with techrepublic premium

Benchmark leader in science, mathematics and codes

The Academic Reasoning Benchmark at Gemini 2.5 Pro Excel, 86.7% on AIME 2025 (Mathematics) and 84.0% on GPQA Diamond Benchmark (Science). In the final examination of humanity – a comprehensive test of thousands of questions in mathematics, science and humanities goes with a score of 18.8%.

In particular, these results were obtained without the use of expensive test-time techniques, which allow models such as O1 and R1 to continue learning during evaluation.

In software development benchmark, Gemini 2.5 Pro performance is mixed. This scored 68.6% on the AIDER polyglot benchmark for code editing, making most of the top-level models better. However, it scored 63.8% on SWE-Bench verified, finished second in Cloud Sonnet 3.7 in comprehensive programming works.

Despite this, google says Create a video game from a prompt,

The model supports a reference window of a million tokens, which means that it can process the 750,000-word prompt, or equal to the first six Harry Potter books. Google has planned to increase this limit to two million tokens in the appointed time.

The Gemini 2.5 Pro is currently available through the Gemini Advanced App, which requires $ 20-Mahine membership, and for developers and enterprises through Google AI Studio. In the coming weeks, Gemini 2.5 Pro will be made available on the machine-learning platform of Google for Vertex AI, developers, and pricing details for various rate limits will also be offered.