Google Announces "Gemini 2.0" — Multimodal AI Built for the Agentic Era
Google announced its new-generation model family, “Gemini 2.0.” The first release, the experimental “Gemini 2.0 Flash,” outperformed 1.5 Pro on key benchmarks while running twice as fast. Google positioned this update as being built “for the agentic era,” clearly signaling a direction where AI autonomously makes use of tools.
Details
- Native multimodal output: The model itself can now directly generate image output and multilingual text-to-speech audio
- Tool use: Equipped with “native tool use,” allowing the model to autonomously call external tools such as Google Search or execute code
- Multimodal Live API: Released a new API for real-time interaction using audio and video
- Availability: Available immediately to developers via the Gemini API, Google AI Studio, and Vertex AI. A chat-optimized version was also available in the Gemini app’s model selection menu
- Phased rollout: Full general availability was planned for January 2025 onward, with this release going first to developers and trusted testers
What happened next
Gemini 2.0 continued its rollout toward general availability, and in March 2025, “Gemini 2.5,” a “thinking model” with enhanced reasoning ability, arrived. The agentic direction set out with Gemini 2.0 became the foundation of Google’s subsequent product strategy, combining search and tool use for autonomous AI applications.