- Google announces Gemini 2.5
- Gemini 2.5 Pro Experimental is available for paid subscribers right now
- Tops Humanity’s Last Exam, the most difficult AI benchmark
Google just announced Gemini 2.5, and it’s a major upgrade to Gemini that the company is calling its “most intelligent AI model” yet.
Announced on the company’s blog, Google revealed the experimental version of 2.5 Pro, which is available today for all Gemini Advanced subscribers. More 2.5 models will arrive in the future.
Google’s Gemini 2.5 models are a new generation of thinking models that are able to reach “a new level of performance by combining a significantly enhanced base model with improved post-training.”
2.5’s thinking capabilities will be implemented into all future Google AI models, which the company says will allow them to “handle more complex problems and support even more capable, context-aware agents.”
So what does this mean? Well, Google is doubling down on its impressively frequent AI updates, and this time we’re getting better reasoning capabilities than ever before.
Available right now, you can access Gemini 2.5 Pro Experimental simply by selecting the model in the Gemini app or directly in Google’s AI Studio. You’ll need a Gemini Advanced subscription to see this as an option.
Pricing for the improved model (for those who want to use it for scaled production use) will be announced in the coming weeks, and more 2.5 models are expected to launch in due course.
Gemini 2.5 Pro is impressive
Google shared some benchmark results for Gemini 2.5 Pro Experimental and the results are seriously impressive.
The new AI model scores 18.8% on Humanity’s Last Exam compared to 14% for ChatGPT‘s o3-mini and 8.6% for DeepSeek R1. Humanity’s Last Exam is the most thorough and difficult AI benchmark, so to score substantially higher than its competitors is no mean feat.
18.8% is the highest score we’ve ever seen on Humanity’s Last Exam (without tool use). Google is calling Gemini 2.5 Pro’s reasoning capabilities “state-of-the-art” and it’s clear to see why.
Google continues to drive forward with its AI development at a rapid pace. Just last week the company made Gemini Deep Research free for all and followed that up with improvements to its impressive AI podcasting tool, NotebookLM.
We’ll be testing Gemini 2.5 Pro and putting the new Experimental model through its paces, so stay tuned to TechRadar for further Google AI coverage.
You might also like
https://cdn.mos.cms.futurecdn.net/MTfPbjwsR63QQUi4mj5UK9-1200-80.jpg
Source link
john-anthony.disotto@futurenet.com (John-Anthony Disotto)
