Google has introduced Gemini 3, which stands out with its multimodal intelligence, advanced reasoning, and coding capabilities. While the model is available for use starting today, the Deep Think mode will be available soon.
A New Standard in AI Intelligence
Starting today, Google has begun rolling out its new AI models, named Gemini 3, to users. The company defines this series as the “smartest” AI systems with high factual accuracy. Gemini 3 is seen as a crucial step for Google to take the lead in consumer-focused AI, especially following the issues with OpenAI‘s GPT-5 launch. The DeepMind team, led by Demis Hassabis and Koray Kavukçuoğlu, describes Gemini 3 as “our smartest model capable of bringing every idea to life.”
Introduced just seven months after Gemini 2.5, this new version stands out as Google’s most capable large language model (LLM) to date. The most notable member of the new series, Gemini 3 Pro, has been opened to all users via the Gemini app starting from the launch day. Additionally, Google AI Pro and Ultra subscribers in the US will be able to experience Gemini 3 Pro via AI Mode within Google Search.
Multimodal Capabilities and Product Integration
Gemini 3 Pro is designed as a naturally multimodal model. This means it can process text, image, and audio data simultaneously. For example, it can take photos of a recipe and turn them into a cookbook or create interactive cards from a video lecture series.
The improvements of the new model will be felt across most Google products. In the Gemini app, users gain a workspace where they can create more comprehensive programs. Additionally, thanks to generative interfaces tested in Gemini Labs, Gemini 3 Pro can design visual and magazine-style content or custom user interfaces. In Search with AI Mode, responses to your queries are presented with visual elements such as images, tables, grids, and simulations.
Smarter, More Direct, and Reliable
Google emphasizes that thanks to the updated “query chunking” technique, the model can better understand the intent of questions and find content missed in previous searches. Google defines Gemini 3 Pro as a more reliable model compared to ChatGPT‘s empty praises. Google DeepMind VP of Product Tulsee Doshi states that answers are now smart, concise, and direct, offering real information instead of clichés and excessive compliments. In other words, the model now engages in much less sycophancy.
Gemini 3 Pro’s reasoning and task completion capabilities have also been improved. It can perform more complex tasks and engage in long-term planning. These features are used in the Gemini Agent experimental feature, which can perform operations such as email review, travel research, and booking on behalf of the user.
Advanced Logic: “Deep Think” Mode
Furthermore, the Deep Think mode, which is currently open only to test users, further increases reasoning capacity. After passing safety tests, Deepthink will be offered to Google AI Ultra subscribers in the coming weeks.
Deep Think mode takes the model’s logic and multimodal understanding capacity even further. In tests, Gemini 3 Deep Think surpasses Gemini 3 Pro with scores of 41% in Humanity’s Last Exam and 93.8% in GPQA Diamond. It also demonstrates its ability to solve new and unique challenges with 45.1% in ARC-AGI-2.
Powerful Tools for Developers: Antigravity
Gemini 3 offers powerful tools for developers. With its zero-shot production capability, it can process complex instructions to create richer and more interactive web interfaces. It achieves an Elo score of 1487 in WebDev Arena, making it the most competent model in agentic coding and vibe coding. It provides a significant advantage over 2.5 Pro in Terminal-Bench 2.0 and SWE-bench Verified tests.
Introduced simultaneously with Gemini 3, the coding interface named Antigravity offers a multi-panel and agent-assisted experience. Working similarly to tools like Warp or Cursor 2.0, Antigravity combines a ChatGPT-style command window, a terminal interface, and a browser window showing the effects of changes. The model is also accessible on third-party platforms such as Cursor, GitHub, JetBrains, Manus, and Replit.
Breaking Records in Benchmarks
Looking at the performance side, the model also shows strong performance in independent measurements. It achieved the highest score of all time with 37.4 points in Humanity’s Last Exam. It also ranks at the top in LMArena, which measures user satisfaction. It set a new standard with 23.4% in the field of mathematics at MathArena Apex. Standing out with its multimodal skills, Gemini 3 Pro offers reliable solutions to complex problems with 81% success in MMMU-Pro, 87.6% in Video-MMMU, and 72.1% in SimpleQA Verified.
Global Availability and The Gemini 3 Era
According to Google, the Gemini app has 650 million monthly active users, and 13 million software developers are actively using the model in their workflows. The distribution of Gemini 3 has spread to a wide ecosystem as of today. While the model is available to all users in the Gemini app, Google AI Pro and Ultra subscribers can also benefit from the new capabilities via AI Mode. Developers can access Gemini 3 through the Gemini API, AI Studio, Google Antigravity, and Gemini CLI. On the corporate side, the model has been integrated into Vertex AI and Gemini Enterprise platforms.
You Might Also Like;
- We Selected 10 Series Similar to Stranger Things for Those Who Love It
- Where and How is Silver Used in Electric Vehicles?
- Hyundai Unveils Its Multi-Purpose Wheeled Robot
-
-
-
-
-
-
-
-
-
-
-
-
