Details
- Google AI Developers introduced Gemini API skills to equip coding agents with up-to-date SDK knowledge, tested on gemini-3.1-pro-preview model.
- Evaluation achieved 96.6% success across 117 coding prompts, covering agentic, chat, and SDK use cases.
- Skills enable agents to handle complex coding tasks using Gemini 3.1 Pro, which supports text, images, audio, video, code, and up to 1 million token context window.
- Developers can integrate via provided links: learn how at one URL, add to workflows at another.
- Ties into Gemini Code Assist, offering AI-powered code completion, generation, chat support in IDEs like VS Code and JetBrains, with free, Standard, and Enterprise editions.
- Model excels in agentic coding, deep reasoning, and problem-solving, available through Gemini API in AI Studio, Vertex AI, and more.
Impact
Google's Gemini API skills push coding agents toward higher reliability, with 96.6% success outpacing typical benchmarks and pressuring rivals like OpenAI's GPT-4o and Anthropic's Claude in agentic coding tasks. This lowers barriers for developers building autonomous workflows, widening AI adoption in software development. By leveraging Gemini 3.1 Pro's million-token context and multimodal capabilities, it narrows the gap with leaders in production-ready code generation, potentially accelerating enterprise shifts to Google Cloud integrations amid growing demand for secure, customizable AI tools.
