Not all models are built to think — this one is

GOOGLE

Not all models are built to think — this one is

Google has just released Gemini 2.5 Pro (experimental), their most intelligent AI model to date.

It builds on everything that made previous models successful, like multimodality and long context, but pushes things further with significantly improved reasoning and coding abilities.

Unlike standard models that respond based on predictions alone, Gemini 2.5 Pro is what Google calls a “thinking model.”

It processes information in steps before answering, reasoning through the problem rather than jumping to answers (A skill I also need to learn).

That shift has helped it score top marks across several challenging benchmarks, especially in maths, science, and software development.

On LMArena, a leaderboard based on human preferences, 2.5 Pro debuts in the #1 spot by a wide margin.

It also performs strongly on tests like GPQA, AIME 2025, and Humanity’s Last Exam, a benchmark designed to test the edge of human knowledge and reasoning.

What’s new with Gemini 2.5:

• Performs well on advanced maths and science benchmarks like GPQA and AIME 2025, without relying on costly add-ons like majority voting.

• Handles complex coding tasks with ease, from generating full web apps to editing and transforming code. Scored 63.8% on SWE-Bench Verified, a key industry benchmark.

• Processes massive volumes of input—up to 1 million tokens now, with 2 million coming soon, across formats like text, audio, video, and full codebases.

It’s giving STEM vibes

In terms of coding, 2.5 Pro is a major step up from its predecessor.

It performs well in real-world developer tasks such as editing, transforming, and generating code.

It even builds functional applications from a single-line prompt, like generating a basic video game.

It scored 63.8% on SWE-Bench Verified, an industry standard coding benchmark.

Gemini 2.5 Pri also benefits from a long context window, currently up to 1 million tokens (2 million coming soon), which allows it to work across massive datasets, including full code repositories and multimedia inputs like audio, video, and images.

The model is now available in Google AI Studio and the Gemini app for Advanced users, with a rollout to Vertex AI expected shortly.

Pricing and higher-rate access for scaled usage will be announced in the coming weeks.

This AI model probably understands my situationship better than I do.