Gemini 3 Officially Introduced as Google’s Most Intelligent Model: Gemini 3 Pro Now Available Across Google Services
Google has now officially introduced Gemini 3 as its most intelligent model, combining all the capabilities of Gemini. It has already been rolled out across several Google services, and is currently accessible.
Check out more about its performance and other details below.
Google’s Gemini 3 – Officially Introduced
The newly-introduced Gemini 3 has been brought out as the best model in the world in terms of multimodal understanding, and it is Google’s most powerful agentic and vibe coding model to date. It is built to offer state-of-the-art reasoning, and the brand has marked the Gemini 3 era with its Gemini 3 Pro release in preview. Additionally, the brand has also released the Gemini 3 Deep Think mode too, for an even enhanced reasoning performance.
Speaking more, the Gemini 3 Pro has topped the LMArena Leaderboard with a 1501 Elo score, and has also attained top scores on Humanity’s Last Exam (37% without tools) and on GPQA Diamond (91.9%). Furthermore, it was also able to achieve a 23.4% score on MathArena Apex too. An 81% score on MMMU-Pro, 87.6% on Video-MMMU, and 72.1% on SimpleQA Verified also showcases its top performance. In the case of Gemini 3 Deep Think, it again outperforms the performance of Gemini 3 Pro. Notably, it has achieved 41.0% on Humanity’s Last Exam (without tools), 93.8% on GPQA Diamond, and 45.1% on ARC-AGI-2 (with code execution).
On the safety side, Gemini 3 is also pointed out as Google’s most secure model yet, ensuring reduced sycophancy, increased prompt injection resistance, and improved protection against cyber attacks. In house testing for critical domains in the Front Safety Framework, partnered evaluations with leading subject matter experts, early access to bodies like UK ASI, obtained independent assets from experts like Apollo, Dreadnode, Vaultis, and more are other security aspects.
Regarding the availability, Gemini 3 Pro is now available on the Gemini app (for everyone) and in AI Mode on Search (for Google AI Pro and Ultra subscribers). For developers, it is available on Gemini API in AI Studio, Antigravity, and on Gemini CLI. It is also accessible on Cursor, JetBrains, GitHub, Replit, Manus, and certain other third-party platforms. Availability on Vertex AI as well as on Gemini Enterprise is also being offered by the brand. Gemini 3 Deep Think mode is currently in tests, and will be rolled out to Google AI Ultra subscribers later in the coming weeks. Also, the brand has promised that more models will be introduced to the Gemini 3 series soon.
Stay tuned for more updates.
