Google has announced the launch of Gemini 3, the latest iteration of its flagship AI model. Building on the success of Gemini 2.5, the new AI model promises enhanced simulated reasoning abilities and improved understanding of text, images, and video content. This advancement brings the tech company closer to achieving artificial general intelligence (AGI), a significant milestone in AI development.
Gemini 3 boasts improved factuality, evident in its performance on various benchmark tests. Scoring 72.1% in the 1,000-question SimpleQA Verified test, the model showcases progress in general knowledge understanding. Additionally, Gemini 3 excels in challenging assessments like Humanity’s Last Exam, scoring 37.5% without tool assistance, demonstrating its prowess in tackling PhD-level knowledge and reasoning.
The model’s math and coding capabilities have also seen enhancements, setting new records in MathArena Apex, WebDev Arena, and SWE-bench Verified. Gemini 3’s ability to generate code, scoring an impressive 76.2% in SWE-bench Verified, highlights its potential in practical applications.
Alongside Gemini 3, Google has introduced Antigravity, an AI-first integrated development environment (IDE). This tool aims to streamline AI development processes, offering developers a comprehensive platform to create innovative AI solutions.
Source: Ars Technica