Google today announced an updated Gemini 2.5 model with several improvements that can be observed in popular AI benchmarks. Google particularly highlighted that the new Gemini 2.5 Preview 06-05 "Thinking" model performs better in coding, math, science, and reasoning.
Our latest Gemini 2.5 Pro update is now in preview.
— Sundar Pichai (@sundarpichai) June 5, 2025
It’s better at coding, reasoning, science + math, shows improved performance across key benchmarks (AIDER Polyglot, GPQA, HLE to name a few), and leads @lmarena_ai with a 24pt Elo score jump since the previous version.
We also… pic.twitter.com/SVjdQ2k1tJ
Last month, during Google I/O, Google released an updated version of the Gemini 2.5 Pro model with significant improvements. Today's update builds on the release of the previous month with further enhancements. In addition to benchmark improvements, based on user feedback, this updated Gemini 2.5 Pro model also comes with improved style and structure, which will result in creative and better-formatted responses for end users.
You can find the benchmark comparison with other leading AI models below.
As you can notice in the above table, this updated Gemini 2.5 Pro preview model is now SOTA (State-of-the-Art) in coding benchmarks like Aider Polyglot. It also scored SOTA performance on GPQA and Humanity’s Last Exam (HLE) benchmarks, which test math, science, knowledge, and reasoning capabilities. In real-world testing, this latest 2.5 Pro model scored 24 points better on LMArena, maintaining its lead, and a 35-point jump on WebDevArena at 1443.
Developers can access this latest Gemini 2.5 Pro preview model through the Gemini API via Google AI Studio and Vertex AI. General consumers will be able to access this model via the Gemini app. Google also confirmed that the Gemini 2.5 Pro model will be generally available in a couple of weeks, allowing developers to start using it in production-ready enterprise-scale applications.
0 Comments - Add comment