Google has officially launched its most advanced AI model yet, the Gemini 3 Pro, and it’s already setting new records across major AI benchmarks. After months of hype, Google says Gemini 3 Pro is part of a new family of “multimodal reasoning models” and marks another step toward its long-term AGI goals.
Gemini 3 Pro uses a Mixture-of-Experts (MoE) design built on the Transformer architecture and was trained fully on Google TPUs, which makes its performance even more impressive. In multiple tests, the model beat out other frontier AI systems. On Humanity’s Last Exam, it scored 37.5%, far higher than OpenAI’s GPT-5.1 at 26.5%. It also topped the LMArena leaderboard and posted strong results in ARC-AGI-2 and SWE-Bench.
Google is also preparing Gemini 3 Deep Think, a more powerful version coming to AI Ultra subscribers. Developers will soon have access to the new Antigravity platform, which lets AI agents plan tasks, write code, and control tools directly.
Gemini 3 Pro is now rolling out in the Gemini app and in AI Mode on Google Search for Pro and Ultra users. The Gemini 3 Pro model has not been officially released locally; only the 2.5 Pro version is available in the Philippines.
0 Comments
Leave a Reply