Google DeepMind Releases Gemini 2.5 Pro with Native Tool Use

TL;DR: Google DeepMind releases Gemini 2.5 Pro with native tool-use abilities, enabling direct code execution, web browsing, and function calling without external orchestration.

Key Points

• Gemini 2.5 Pro includes native tool use: function calling, code execution, and web search are built into the model.
• The model can autonomously decide when to call tools, execute code, or search the web during inference.
• Benchmarks show 25% improvement over Gemini 2.0 on agentic tasks requiring multi-step tool use.
• Available through Google AI Studio and Vertex AI, with pricing at $3.50 per million input tokens.
• Google is also releasing an open-weights version (Gemini 2.5 Flash) for on-device and edge deployments.

Why It Matters

Native tool use removes the need for external orchestration frameworks to build AI agents. This could simplify the architecture of many AI applications and reduce latency in agentic workflows. The open-weights Flash variant also opens doors for developers who need to run models locally or in resource-constrained environments.

Source: Google AI Blog