Google releases ‘most intelligent model to date,’ Gemini 2.5 Pro

After a few months after the Gemini 2.0 and rise Depth, Google He still announced the “smartest model”, twins 2.5, with meditation and better performance and accuracy.

Gemini 2.5 comes in three months after the smartest model released his family, The Gemini Present 2.0 Reasons for justification and agent. This new model is available in Google AI Studio as a Gemini 2.5 Pro (experimental) and is available for advanced users of the Gemini in the Twins chat interface. Will soon be available in Vertex AI.

Koray Kavukcuoğlu, Google Deepmind in CTO, said Writing a blog These twins 2.5 represents the next step in Google’s “EU is smart and able to think.”

“Now, with the Gemini 2.5, we have achieved a new performance level by combining a significantly developed base model with advanced post-education,” Kavukcuoglu wrote. “We are moving forward, we directly build the ability to directly models, so they can handle more complex problems and support more skilled, contextual agents.”

More context and understanding

Twins 2.0 and Twins 2.0 Flash ThinkingBefore the Twins 2.5 Pro “thinks”. The new model can manage multimodal access from text, audio, pictures, videos and large data staff. Gemini 2.5 Pro can also understand all code deposes for coding projects.

Gemini 2.5 Pro offers the largest context windows available for experimental models over twins. 1 million Token is sent with context window, but will soon be expanded to 2 million token. Google Ai Studio Product Manager Logan Sented in Kilpatrick X This Gemini 2.5 Pro “is the first experimental model with higher grade restrictions + calculation.”

Google will soon plan to release prices for Gemini 2.5 models.

Advanced coding and justification performance

Google said that the model has taken advanced thoughtful criteria tests. The company Said Gemini 2.5 Pro is headed by matches and science benchmarks like GPGA and AIME 2025. ” Kavukcuoğlu, the model, in the final examination of the “humanity”, is most likely to seize the most 18.8% between models, “human knowledge and thought.

Gemini 2.5 Pro is carried out in coding tasks in the same time, and in special criteria in special criteria, twins are better than 2.0. Google, along with visual attractive web applications and agent code apps, code transformation and editing, the new model “Excel”.

https://www.youtube.com/watch?v=rlcbspgos6s

A more competitive market

Gemini 2.5 Pro in December, in December, the Gemini enters an important environment in an important environment than 2.0. Deepseek’s Reasoning Language Model (LLM) DeepSeek-R1’s release showed that strong models can Make good performing with a fraction Training and calculation cost. In addition, DeepSeek showed that Open source models can compete with more indoor welding llms OpenO1 and O3 models.

Along with the expanding model offers from DeepSseek, Google must compete with Openai’s justification models. Although the newest model Openai was GPT-4.5 – A reasoning model – the company is still More justification is expected to be developed Models soon.

Twins 2.5 Google’s second new model this month. In March, the company published its latest version Small Language Model, Gemma 3128,000 was the best for use in a Token context model and use on go-to-go devices.