Google has aggressively rolled out Gemini in the past year, releasing the much-improved Gemini 2.5 family and introducing variants of the model into Search, Gmail, and virtually everything else the company makes.
Now Google's increasingly inevitable artificial intelligence is getting an update. Gemini 3 Pro available in limited form todaywith more exciting visuals and fewer lies, Google says. The company also says Gemini 3 sets a new high-water mark in Vibe coding, and Google announces a new AI-focused integrated development environment (IDE) called Antigravitywhich is also available today.
The first member of the Gemini 3 family.
Google says the release of Gemini 3 is another step towards artificial general intelligence (AGI). The new version of Google's flagship artificial intelligence model expands its ability to model reasoning and improve understanding of text, images and video. Testers are loving it so far, with Google's latest LLM once again topping the LMArena leaderboard with an ELO score of 1501, beating the Gemini 2.5 Pro by 50 points.
Factuality has been an issue for all AI models, but Google says Gemini 3 is a big step in the right direction, and there are plenty of tests that can tell the story. In the 1,000-question SimpleQA Verified test, Gemini 3 scored a record 72.1 percent. Yes, this means that the modern LLM program still doesn't answer nearly 30 percent of common questions, but Google says it still shows significant progress. On the much more difficult “Last Exam of Humanity,” which tests doctorate-level knowledge and thinking, Gemini set another record, scoring 37.5 percent without using instruments.
Math and programming are also the focus of Gemini 3. The model set new records in MathArena Apex (23.4 percent) and WebDev Arena (1,487 ELO). In the SWE-bench Verified test, which tests a model's ability to generate code, Gemini 3 scored an impressive 76.2 percent.
