Skip to main content
Background Image

Google Launches Gemini 3: Its Most Intelligent AI Model Yet

·323 words·2 mins·
Pini Shvartsman
Author
Pini Shvartsman
Architecting the future of software, cloud, and DevOps. I turn tech chaos into breakthrough innovation, leading teams to extraordinary results in our AI-powered world. Follow for game-changing insights on modern architecture and leadership.

Google has officially launched Gemini 3, marking what the company calls “a new era of intelligence.” The announcement, made by CEO Sundar Pichai alongside Google DeepMind leadership, positions Gemini 3 as Google’s most capable AI model, combining advanced reasoning with multimodal understanding.

Benchmark-Topping Performance
#

Gemini 3 Pro arrives with impressive benchmark results. The model tops the LMArena Leaderboard with a breakthrough score of 1501 Elo and demonstrates PhD-level reasoning with 37.5% on Humanity’s Last Exam (without tools) and 91.9% on GPQA Diamond. In mathematics, it achieves 23.4% on MathArena Apex—a new state-of-the-art for frontier models.

The multimodal capabilities are equally strong, with 81% on MMMU-Pro and 87.6% on Video-MMMU. Google also highlights 72.1% on SimpleQA Verified, indicating improved factual accuracy.

Deep Think Mode
#

Beyond the standard model, Google introduces Gemini 3 Deep Think—an enhanced reasoning mode that pushes performance further on complex problems. In testing, Deep Think achieves 41.0% on Humanity’s Last Exam and 93.8% on GPQA Diamond, outperforming the already-impressive base model.

Deep Think also scores 45.1% on ARC-AGI-2, demonstrating its ability to tackle novel challenges. Google is providing early access to safety testers before rolling it out to Google AI Ultra subscribers.

Broad Availability
#

Gemini 3 is rolling out across Google’s product ecosystem from day one—a first for the company. Users can access it through the Gemini app, AI Mode in Search, AI Studio, and Vertex AI. Developers also get access through Google’s new agentic development platform, Google Antigravity, and the Gemini CLI.

The model excels at “vibe coding” and agentic tasks, topping the WebDev Arena leaderboard at 1487 Elo and scoring 76.2% on SWE-bench Verified. It’s also available in third-party platforms including Cursor, GitHub, JetBrains, Replit, and others.

Looking Ahead
#

Google describes Gemini 3 as smarter, more concise, and less sycophantic—“telling you what you need to hear, not just what you want to hear.” Additional models in the Gemini 3 series are planned for release soon.

Related

Google's Gemini 3 Pro Joins GitHub Copilot Model Lineup
·332 words·2 mins
Google Brings Gemini to Fitbit: Your AI Health Coach Is Here
·334 words·2 mins
Google Replaces Assistant with Gemini for Home: Smarter, But at What Cost?
·317 words·2 mins