Google unveils 'its most intelligent reasoning AI yet'
Where does it really stand out? Memory.
AI is getting smarter, but not in the way you might think. Speed isn’t the problem— accuracy is. AI models have been great at generating words and code, but they often get things wrong.
Now, Google says it’s found a fix with its "most intelligent AI model yet”– Gemini 2.5. Unveiled on Tuesday, Gemini 2.5 is part of a new wave of AI reasoning models designed to pause and reason before answering. Unlike traditional AI that spits out responses instantly, these models take extra time and computing power to verify facts, process logic, and think through problems—kind of like how a human would before making an important decision.
To kick things off, Google is rolling out Gemini 2.5 Pro Experimental, a version fine-tuned for complex reasoning. While past AI models prioritized speed, this reasoning model focuses on handling bigger, messier problems that require actual thought rather than simple pattern-matching.
Kelechi Edeh
While Google has played with AI reasoning before—like the Gemini 2.0 Flash Thinking, which was first released in December—this 2.5 model is an improvement and its most serious attempt yet to leapfrog offerings from the competition.
Video Credit: Google
This shift to reasoning models isn’t just a Google thing. OpenAI kicked off the AI reasoning race last year with its o1 model, and since then, companies like Anthropic, DeepSeek, and xAI have been scrambling to develop their own versions.
So, how does Gemini 2.5 compare? Google claims it outperforms OpenAI’s GPT-4.5, Claude’s Sonnet 3.7, and DeepSeek’s R1 in several benchmarks, including Humanity’s Last Exam (HLE), GPQA diamond and AIME 2025.
Where it really stands out, though, is memory. It can process 1 million tokens in a single session—roughly 750,000 words, longer than The Lord of the Rings trilogy. And soon, Google says it’ll double that to 2 million tokens, allowing for deeper, more complex conversations.
Going forward, Google says it plans to bake reasoning into all future Gemini models, making them better at tackling complex tasks and supporting AI agents that don’t just follow instructions—they figure things out.
The experimental 2.5 model is available now in Google AI Studio and $20/month for Gemini Advanced subscribers, while other versions of the 2.5 are expected later. Details about API pricing are still unclear, but Google says more details are coming soon.
