Google launches Gemini 3: surpasses GPT-5.1 on many important AI tests
Gemini 3 officially launched with stronger inference capabilities, leading many AI benchmarks and surpassing GPT-5.1 in key assessments. Deep Think mode is also greatly improved and ready for both consumers and businesses.
Google has just officially announced Gemini 3 – the AI model that has been awaited for a long time. After a series of 'teasers', Google DeepMind has officially unveiled the new generation AI platform, possessing the ability to reason and process multimedia that is described as the most outstanding ever.
As announced, Gemini 3 is currently leading the LMArena leaderboard with a new record Elo score of 1501. The model also scored 37.5% in Humanity's Last Exam, 91.9% in GPQA Diamond, and 23.4% in MathArena Apex, all of which are top performance benchmarks. Multimodality capabilities have also been significantly improved: Gemini 3 Pro scored 81% in MMMU-Pro and 87.6% in Video-MMMU. In the SimpleQA Verified accuracy test, the model scored 72.1%, topping the leaderboard.
Google also shared a full comparison chart between Gemini 3 Pro, GPT-5.1, and Claude Sonnet 4.5. In parallel, the company introduced Gemini 3 Deep Think mode, a version optimized for complex inference tasks. Gemini 3 Deep Think scored 41% on Humanity's Last Exam, 93.8% on GPQA Diamond, and 45.1% on ARC-AGI-2 (with code execution support, ARC Prize Verified version). Despite the huge performance increase, the entire Gemini 3 family maintains a context window of up to 1 million tokens.
Unlike previous generations that took months to reach users, this time Google chose to launch the product faster and more widely. AI Mode in Google Search used Gemini 3 to bring a content-generated search interface, including a more intuitive layout, interactive simulations, and dynamic tools that appear immediately according to user queries.
On the SWE-bench Verified test for programming ability, Gemini 3 Pro scored 76.2% – slightly lower than GPT-5.1 and Sonnet 4.5. However, Google says Gemini 3 is available to developers through Google AI Studio, Vertex AI, Gemini CLI, Cursor, GitHub, JetBrains, Manus, Replit, and the company's new agentic development platform, Google Antigravity.
For general users, Gemini 3 is available in the Gemini app. Google AI Pro and Ultra subscribers can experience Gemini 3 in AI Mode on Google Search. For businesses, the model is available through Vertex AI and Gemini Enterprise. Gemini 3 Deep Think will be available to Google AI Ultra customers in the coming weeks.
- Google Launches Gemini CLI, Bringing Gemini to the Terminal
- Google launches Gemini 2.5 Deep Think - surpassing OpenAI o3 and Grok 4 in performance
- Tips for quickly creating automated tests on Gemini
- Google Launches Gemini CLI GitHub Actions: Automate PR Review, Issue Triage, and More
- 4 Google apps that work better with Gemini.
- 4 Useful Ways to Use Gemini AI in Google Slides
- What is Google Gemini? How does Gemini work?
- Why using Google Keep with Gemini really makes a difference?
- Instructions for getting started with Gemini Embedding 2
- How to integrate Gemini with Google Calendar: Master your schedule with AI in 30 seconds.
- Google launches Gemini 3 Pro Image: next-generation imaging model with higher accuracy
- How to use Gemini 3.1 Pro Preview in Google AI Studio
- Google Gemini Launches Awesome Comic Drawing Feature: Just Enter Your Idea, AI Will Take Care of the Rest
- Gemini is no longer as useless as it was two years ago.