(Image source from: x.com/Google)
Google has finally launched the Gemini 3 series of artificial intelligence (AI) models on Tuesday. The tech giant from Mountain View claims this is their smartest AI model yet, emphasizing that it surpasses its previous version along with OpenAI's GPT-5.1 in every important test. The updated AI model shows enhancements in several areas, such as logic, dialogue, programming, math, and independent capabilities. The company states that Gemini 3 represents a significant advancement in creating complex interactive experiences for users. In a blog entry, the tech giant announced the debut of the new Gemini 3 AI models. This release follows several vague posts about it from numerous Google executives on X, which used to be known as Twitter. The lineup consists of two models: Gemini 3 Pro and Gemini 3 Deep Think. The company pointed out that these models are being widely released. This means users can explore their features through Gemini apps and websites, the AI Mode in Search, AI Studio, and Vertex AI for developers, along with a new development platform called Google Antigravity.
Reliance Jio has also revealed that all of its users who signed up for the free Google AI Pro plan will gain access to Gemini 3 Pro as well. However, not all these models will be accessible to everyone. The company is in the process of previewing Gemini 3 Pro, which is the most sophisticated version in the series. This will be the version used across Google's entire range of products. On the other hand, Gemini 3 Deep Think, which is the company's upgraded reasoning model, is currently only available to safety testers. Even after its official launch, this model will be limited to Google AI Ultra subscribers. Regarding its performance, the company shared the benchmark results for Gemini 3 Pro from internal testing. According to the information provided, Gemini 3 Pro exceeds its previous model and OpenAI's GPT-5.1 in all main benchmarks. It also outperforms Claude 4.5 Sonnet in nearly all tests except for two (AIME 2025 and SWE-Bench for math and coding, respectively).
Gemini 3 Pro achieved a score of 37.5 percent on the Humanity's Last Exam, which tests academic reasoning, without any tools. This test is regarded as the toughest for AI models, and the previous highest score was 26.5 percent from GPT-5.1. In addition, it received 31.1 percent on the ARC-AGI-2 benchmark, which involves visual reasoning puzzles and is a standard for artificial general intelligence (AGI). Before this, GPT-5.1 had the record at 17.6 percent, while Gemini 2.5 Pro only managed 4.9 percent on the same benchmark. The company claims that Gemini 3 Pro also makes great improvements in coding, particularly in frontend development. They state that this AI model can now create fully functional websites, applications, and scalable vector graphics (SVGs).
Regarding Gemini 3 Deep Think, the company asserts that it surpasses 3 Pro's capabilities, reportedly scoring 41.0 percent on Humanity's Last Exam and 45.1 percent on ARC-AGI-2. However, this version is not yet available for users to access.
Additionally, the company launched Google Antigravity, a new platform for developers that enables them to work with AI agents. It is an integrated development environment (IDE) powered by AI that allows users to utilize the reasoning, tool abilities, and coding skills of Gemini 3. The agents operate in a dedicated isolated environment, which includes an editor, terminal, and browser. This setup enables agents to independently organize and perform complicated software tasks on behalf of users while also checking their own programming. Antigravity includes the company's newest Gemini 2.5 Computer Use model for web browsing and Nano Banana, in combination with Gemini 3 Pro.
