We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
DeepMind Blog·Nov 18·5 min read
A new era of intelligence with Gemini 3
TL;DR
AI models have struggled with reasoning and multimodal tasks, limiting their effectiveness. Google has developed Gemini 3, an advanced AI model that enhances these capabilities.
✦ Why It Matters
Engineers can leverage Gemini 3's advanced reasoning and multimodal capabilities to enhance their AI applications.
Key Takeaways
- Explore Gemini 3's capabilities in the Gemini app for enhanced learning.
- Utilize Gemini 3 in AI Studio for advanced coding projects.
- Leverage Deep Think mode for tackling complex problems.
- Integrate Gemini 3 into your workflows for improved productivity.
Full Summary
Gemini 3 represents a significant advancement in AI, combining enhanced reasoning and multimodal capabilities to assist users in various tasks. It outperforms its predecessor, Gemini 2.5 Pro, in key benchmarks, achieving top scores in reasoning tests and multimodal understanding.
For instance, it scored 1501 Elo on the LMArena Leaderboard and demonstrated PhD-level reasoning on complex assessments. Gemini 3 is designed to help users learn by synthesizing information across text, images, and code, and it can generate interactive tools for educational purposes.
Additionally, the introduction of the Gemini 3 Deep Think mode further enhances its reasoning capabilities, allowing it to tackle even more complex problems. This model is now available in Google products like the Gemini app and AI Studio, making advanced AI accessible to a broader audience.