How good is Google's free AI?

Over the weekend, Google made a surprise announcement that it was making a beta version of the company's newest and smartest model, the Gemini 2.5 Pro, available to all Gemini users. Previously, the model was only available to users of the Gemini Advanced plan, which costs $20 /month.

Gemini 2.5 Pro integrates the thinking capabilities instead of offering them separately. Although it is a beta version, this model already supports some features like apps/extensions, file uploads, and even Canvas.

In the prestigious AI ranking LMArena, Gemini 2.5 Pro is leading. This model also achieved the best results in math (AIME 2025) and science (GPQA diamond) tests without the need for expensive testing techniques.

The trial version of Gemini 2.5 Pro is available to all users for free. Photo: Google.

Breakthrough in programming

Google says that Gemini 2.5 Pro is a “quantum leap” from version 2.0. The model scored 18.8 percent, the highest score among models without external tools, on Humanity’s Last Exam, a dataset designed by hundreds of experts to measure the limits of human knowledge and thinking.

This version also excels at creating intuitive web applications and automated coding, scoring 63.8% on the SWE-Bench automated programming test, much higher than Gemini 2.0 and prominent competing models such as OpenAI o3-mini, Grok 3 Beta.

Gemini 2.5 Pro's benchmarks compared to the smartest chatbots today. Photo: Google.

In addition, the Gemini 2.5 Pro context window is already very large with 1 million tokens, equivalent to 700 pages of documents, and is about to be doubled. The second place is Claude with 200,000 tokens.

This means the model can make sense of huge datasets and handle complex problems from a variety of information sources, including text, audio, images, video, and even source code repositories.

With such a large context window, the model's processor can contain almost the entire source code of a small software project, according to Theo Browne, YouTuber and CEO of Ping Labs.

Content creators challenged the Gemini 2.5 Pro with many difficult tasks such as creating websites and games with just one command.

Multi-application integration

Not only supporting programmers, Gemini 2.5 Pro will become a powerful tool for content creators and designers. The model does better with video, audio, and image creation tasks, and also integrates Canvas, a feature that helps AI draw diagrams, design UIs, and suggest concept ideas.

The most common application of the model is probably in the field of education with the ability to store large amounts of data and process multiple documents without interruption. Students can personalize their learning path, as well as create vivid images and videos for presentations.

Gemini 2.5 Pro is now available on gemini.google.com and Google AI Studio. In the near future, the model will be integrated into Vertex AI, Google Cloud's premium platform that helps businesses deploy and manage AI models at scale.

In Vietnam, users are debating Gemini's programming capabilities. Mr. Doan Kien, founder of Seodo Media Company, believes that Gemini 2.5 Pro is currently in the leading position. However, his article has received many opposing comments, saying that Claude deserves that position.

Comparing the capabilities of AI chatbots. Photo: Peter Yang/X

In the coming time, Google will focus on improving the programming capabilities of Gemini 2.5 Pro. Although there is no end date for the trial version or announced pricing policy, many users predict that in the official phase, Gemini 2.5 Pro will have a higher cost of use than other models, but will come with many flexible options.

Source: https://znews.vn/ai-mien-phi-cua-google-tot-den-dau-post1542121.html