Google has kicked its Gemini rollout into high gear over the past year, releasing the much-improved Gemini 2.5 family and adding various flavors of the model to Search, Gmail, and everything else the company makes.
Now, Google’s increasingly indispensable AI is getting an upgrade. Google says Gemini 3 Pro is available today in a limited form, with more immersive, visual output and less lag. The company also says the Gemini 3 Vibe sets a new high-water mark for coding, and Google is announcing a new AI-first integrated development environment (IDE) called AntiGravity, which is also available today.
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 AI model has expanded simulated reasoning capabilities and shows better understanding of text, images and video. So far, testers like it—Google’s latest LLM once again tops the LMArena leaderboard with an ELO score of 1,501, besting Gemini 2.5 Pro by 50 points.
Factfulness has been a problem for all General AI models, but Google says Gemini 3 is a huge step in the right direction, and there are myriad benchmarks for storytelling. In the 1,000-question SimpleQA verified test, the Gemini 3 scored a record 72.1 percent. Yes, that means the state-of-the-art LLM still solves only about 30 percent of the general knowledge questions, but Google says it still represents substantial progress. On the much more difficult Humanities Final Exam, which tests PhD-level knowledge and reasoning, Gemini set another record by scoring 37.5 percent without tool use.
Math and coding are also a focus of Gemini 3. The model set new records in MathArena Apex (23.4 percent) and WebDev Arena (1487 ELO). In SWE-Bench Verify, which tests a model’s ability to generate code, the Gemini 3 achieved an impressive 76.2 percent.