Gemini 3 - Google's next-generation multimodal understanding and inference AI model

What is Gemini 3?

Gemini 3 is Google's latest AI model, hailed as the world's most advanced multimodal understanding and reasoning model. It boasts powerful reasoning capabilities, breaking numerous benchmark records, such as topping the LMArena Leaderboard with a score of 1501 Elo. Gemini 3 supports multimodal input, including text, images, and video, and can handle complex problems, providing reliable answers. The model incorporates a deep thinking mode, further enhancing its ability to solve complex problems. Gemini 3 can be used for learning and knowledge acquisition, helping developers build applications efficiently.

Users can now build using Gemini 3 in Google AI Studio, Vertex AI, Gemini CLI, and Google's new agent development platform, Google Antigravity. Models are supported on third-party platforms such as Cursor, GitHub, JetBrains, Manus, and Replit, providing developers with a wide range of options for building and developing applications.

Main functions of Gemini 3

• Strong reasoning abilityThe Gemini 3 Pro achieves top-tier inference capabilities, breaking multiple benchmark records, such as topping the LMArena Leaderboard with a score of 1501 Elo, demonstrating doctoral-level ability to solve complex problems.
• Multimodal understandingIt supports multiple modal inputs such as text, images, and videos, achieving high scores of 81% and 87.6% in MMMU-Pro and Video-MMMU tests respectively, and can parse complex charts and dynamic video streams.
• Deep thinking modeThe Gemini 3 Deep Think mode further enhances reasoning capabilities, demonstrating stronger ability to solve complex problems.
• Learning and Knowledge AcquisitionIt helps users learn new knowledge, such as interpreting handwritten recipes, generating interactive learning tools, supporting the analysis of video content, and generating training plans.
• Development and ConstructionAs Google's most powerful programming model, it supports zero-shot generation and complex hint processing, significantly improving development efficiency.
• Planning and Task ManagementThe agent's capabilities have been significantly improved, enabling it to perform long-term planning and task management.
• A brand new development experienceIt integrates with the Google Antigravity platform to automate end-to-end software development, supporting development on multiple platforms such as Google AI Studio and Vertex AI.
• Safety and Reliability: Undergo a comprehensive security assessment, reduce sycophantic behavior, enhance resistance to instant injection attacks, improve cyberattack protection capabilities, and ensure factual accuracy.

Gemini 3 performance

• Excellent reasoning skillsThe Gemini 3 Pro topped the LMArena Leaderboard with 1501 Elo points, demonstrating doctoral-level reasoning capabilities, such as a score of 37.5% on the "Human Ultimate Test" and 91.9% on the GPQA Diamond test.
• Leading in multimodal understandingIt achieved high scores of 81% and 87.6% in the MMMU-Pro and Video-MMMU tests, respectively.
• Breakthrough in Deep Thinking ModeThe Gemini 3 Deep Think mode scored 41.0% in the "Ultimate Human Test", 93.8% in the GPQA Diamond test, and 45.1% in the ARC-AGI-2 test, significantly improving the ability to solve complex problems.
• Outstanding mathematical abilityAchieving a state-of-the-art score of 23.4% in the MathArena Apex test, it sets a new standard for cutting-edge models in mathematical reasoning.
• Improved factual accuracyAchieving a score of 72.1% in the SimpleQA Verified test demonstrates a significant improvement in providing accurate information.
• Development efficiency significantly improvedIt topped the WebDev Arena leaderboard with 1487 Elo points, significantly improving developer efficiency and supporting complex web UI and application development.
• Enhanced tool usage abilityIt scored 54.2% in the Terminal-Bench 2.0 test and significantly outperformed its predecessor in the SWE-bench Verified test, demonstrating excellent performance.
• Long-term planning capability enhancementIt topped the Vending-Bench 2 test, demonstrating excellent long-term task planning and decision-making consistency.

How to use Gemini 3

• Regular usersUse it directly with Gemini, or experience it in Search AI mode on Google AI Pro and Ultra subscription services.
• DevelopersDevelop using Google AI Studio, Vertex AI, Gemini CLI, or Google's new intelligent agent development platform, Google Antigravity.
• Enterprise usersAccess via the Vertex AI platform or Gemini Enterprise Edition.
• Deep thinking modeIn the coming weeks, Google AI Ultra subscribers will be able to use the Deep Thinking mode on Gemini 3, which is currently undergoing a safety evaluation.

Gemini 3 product pricing

Gemini 3.0 Pro introduces a tiered pricing mechanism based on context length, as follows:

• Tasks with less than 200k tokens:
  • Input price: $2.00 per million tokens.
  • Output price: $12.00 per million tokens.
• Tasks with over 200k tokens:
  • Input price: $4.00 per million tokens.
  • Output price: $18.00 per million tokens.

Application scenarios of Gemini 3

• Learning and EducationThe model can integrate multimodal information to generate interactive learning tools, helping users learn new knowledge efficiently.
• Development and ProgrammingAs a powerful programming model, it supports zero-shot generation and complex prompt processing, significantly improving development efficiency.
• Task planning and managementGemini 3's Agent capabilities support long-term task planning, helping users manage complex tasks and daily affairs.
• Content creationGemini 3 can generate high-quality creative content, such as poetry, stories, and game code, helping to express creativity.
• Knowledge Management and SearchProvides a smart, generative UI in Google Search to help users acquire and integrate information more efficiently.