Gemini 3 for developers: New reasoning, agentic capabilities
4 days ago
- #AI
- #Gemini 3
- #Coding
- Gemini 3 Pro is introduced as Google's most intelligent model, surpassing previous versions in AI benchmarks and coding tasks.
- Priced at $2/million input tokens and $12/million output tokens, it's available via Gemini API in Google AI Studio and Vertex AI.
- Gemini 3 Pro excels in agentic coding, scoring 54.2% on Terminal-Bench 2.0, and integrates with tools like Google Antigravity, Android Studio, and more.
- Google Antigravity is a new agentic development platform that enhances developer productivity by managing autonomous agents across workspaces.
- Gemini 3 introduces a client-side bash tool for shell commands in agentic workflows and combines Google Search grounding with structured outputs.
- Vibe coding is enhanced, allowing natural language to translate into fully interactive apps with a single prompt.
- Gemini 3 Pro leads in multimodal understanding, setting records in MMMU-Pro for image and video reasoning.
- Improved spatial reasoning supports embodied tasks in autonomous vehicles, XR devices, and robotics.
- Video reasoning capabilities include high-frame-rate understanding and long-context recall for continuous footage.
- Developers can integrate Gemini 3 Pro into applications via Google AI Studio and Vertex AI, with new API features for deeper reasoning.