Google Launches an Improved Preview of Gemini 2.5 Pro AI Model Which the Company Claims is Better at Coding
Google is making an “improved preview” of its latest AI model Gemini 2.5 Pro available, which the company claims is better at coding. The upgrade is currently accessible through Google AI Studio, Vertex AI, and the Gemini app. The broader rollout will follow in a few weeks.
“Updated preview”
Google announces the “updated version” of Gemini 2.5 Pro. This model builds upon the version Google showcased during the I/O event in May. The new model scores higher on multiple benchmarks. In LMArena, the model rises 24 points to an Elo score of 1470, and on WebDevArena, it achieves an increase of 35 points to 1443. According to Google, it maintains the highest ratings.
The model also scores strongly on coding benchmarks such as Aider Polyglot, and on content-related tests like GPQA and Humanity’s Last Exam. These evaluate, among other things, the reasoning ability and knowledge of AI models in domains such as mathematics and science.
read also
Gemini 2.5: Google Unveils New Generation of AI Reasoning Models
Based on previous feedback, the 2.5 Pro version now provides better structured and more creative responses. According to Google, this makes the model more suitable for applications where clarity and consistency in output are important.
Multiple platforms
Developers can now test the new model via the Gemini API in Google AI Studio and Vertex AI. Both environments now also offer a so-called thought budget functionality. This gives developers more control over costs and computation time when deploying the model.
Additionally, the preview version is also being rolled out in the Gemini app, allowing users to already familiarize themselves with the new capabilities. The stable version of Gemini 2.5 Pro will follow later this month.