Google launches Gemini 2.5 Pro, its ‘most intelligent AI model’ yet

By Judy Sanhz Published March 25, 2025

In a blog post today, Google announced Gemini 2.5 Pro (experimental) for developers and Advanced subscribers, aiming to help you tackle increasingly complex problems. It’s the first in the family and set up to “think” before it speaks.

Google says it’ll be available today in Google AI Studio (its developer platform) and for Advanced subscribers, with Vertex AI support coming soon. Google also claims to outperform the competition, and that Gemini 2.5 Pro takes the number one spot on the LMArena leaderboard with 18.8%, surpassing other AI models such as ChatGPT and Deepseek.

Recommended Videos

“Gemini 2.5 Pro is state-of-the-art across a range of benchmarks requiring advanced reasoning. Without test-time techniques that increase cost, like majority voting, 2.5 Pro leads in math and science benchmarks like GPQA and AIME 2025. It also scores a state-of-the-art 18.8% across models without tool use on Humanity’s Last Exam, a dataset designed by hundreds of subject matter experts to capture the human frontier of knowledge and reasoning.”

If you’re already subscribed to Gemini Advanced, you can use Gemini 2.5 Pro at no extra cost, but if you need to use it at a higher scale, Google will release pricing in the coming weeks for the higher usage. With 2.5 Pro, you get a one-million context window, but a two-million window is coming soon. Google also says Gemini 2.5 Pro can handle difficult problems from sources like audio, text, images, entire code repositories, and video.

Google has kept the AI ball rolling since it launched the “thinking” version of the AI model in December. Recently, Gemini has been available on Google Maps, allowing you to talk to Maps. Google has also announced that Gemini will replace Google Assistant in the coming months.

Computing Writer

Judy Sanhz is a Digital Trends computing writer covering all computing news. Loves all operating systems and devices.

Topics

Computing

The Mac Pro nearly received an M3 Extreme chip twice as powerful as M3 Ultra

High production costs likely killed Apple’s M3 Extreme plans

Apple's Mac Pro on a table at a press event.

Apple discontinued the Mac Pro earlier this year, ending a 20-year run for a computer that once represented the very best of the company’s desktop lineup. However, Apple reportedly had much bigger plans for the machine before ultimately replacing it with the Mac Studio.

According to Bloomberg’s Mark Gurman, Apple developed an M3 Extreme chip that could have offered twice as many CPU and GPU cores as the M3 Ultra. The processor was intended to sit above the Ultra tier and could have finally given the Mac Pro the performance advantage it badly needed. Apple eventually abandoned the chip due to concerns over production costs and limited demand for such an expensive machine.

Computing

Hidden prompts can secretly rewrite an AI’s memory, and researchers say that’s a serious problem

Researchers discover AI attack that rewrites an assistant's long-term memory

Chatbot on a smartphone.

Large language models are getting better at remembering us. Whether it's your preferred writing style, recurring tasks, shopping habits or project deadlines, AI assistants are increasingly storing long-term memories to make future conversations feel more personal and useful. But according to new research, that same feature could become one of AI's biggest security vulnerabilities.

Researchers from New Mexico State University have demonstrated a new attack called GhostWriter, capable of secretly planting false memories inside AI agents. Rather than stealing information outright, the attack manipulates what an AI remembers, potentially causing it to make dangerous decisions long after the original attack has taken place.

Computing

This experiment shows how easy it is to poison an open-weight AI model for under $100

This research raises new doubts about trusting open weight AI models.

Computer, Electronics, Laptop

Open-weight AI models have been having a moment lately. Just this month, Moonshot's massive Kimi K3 model landed close behind Claude Fable 5 and GPT 5.6 Sol in several benchmarks, all while remaining fully open-weight and downloadable by anyone.

However, Katie Paxton-Fear, a cybersecurity lecturer at Manchester Metropolitan University and staff security advocate at Semgrep, managed to poison an open-weight model and proved how easily that openness can be turned against you (via The Register).