Google has made its latest artificial intelligence models, Gemini 2.0, available to the public today, expanding access to its most powerful AI technology, catching up to the rivals like ChatGPT and DeepSeek.
Gemini 2.0 Pro Experimental, Flash and Flash-Lite models features
The release includes three distinct models: Gemini 2.0 Flash for general use, 2.0 Pro Experimental for advanced coding and complex tasks, and 2.0 Flash-Lite for cost-efficient applications. All models are now accessible through the Gemini app, Google AI Studio, and Vertex AI platforms.
Gemini 2.0 Pro Experimental, the company’s most capable model to date, features a massive 2-million-token context window—enough to process approximately 1.5 million words at once. The model can integrate with Google Search and execute code directly, positioning it as a strong competitor in the AI development space.
The standard 2.0 Flash model, which became generally available today after a limited December release, offers improved performance across key benchmarks and will soon include image generation and text-to-speech capabilities. It maintains a 1-million-token context window, making it suitable for processing large amounts of information.
Google also introduced 2.0 Flash-Lite, a new cost-effective option that matches the speed and pricing of its predecessor while delivering superior performance. The model costs developers 0.75 cents per million tokens for text, image, and video inputs, compared to Flash’s 10 cents per million tokens.
These releases align with Google’s broader strategy to develop more “agentic” AI models—systems capable of understanding complex tasks and taking action on users’ behalf.
Google’s new ‘thinking’ Gemini model
Another of Gemini 2.0 models is the experimental Gemini 2.0 Flash Thinking model widely available through the Gemini app. There’s an variant of it that works with apps, like YouTube, Maps and Search. The model sets new performance benchmarks with a 73.3% score on the American Invitational Mathematics Examination (AIME) and 74.2% on the GPQA Diamond science benchmark.
Unlike traditional “black box” AI models, Flash Thinking shows its work by explaining its reasoning process, making it easier for users to understand and verify its conclusions. The model includes native code execution capabilities and features improved reliability with reduced contradictions between its reasoning process and final answers, according to Jeff Dean, chief scientist at Google DeepMind.
Google emphasized its commitment to safety, noting that the Gemini 2.0 lineup incorporates new reinforcement learning techniques that use self-critique to improve response accuracy and handle sensitive prompts more effectively. The company also employs automated red teaming to assess security risks, particularly those related to indirect prompt injection attacks.