These updates are part of Google's broader strategy to refine AI capabilities through experimental releases before wider adoption.

Google has unveiled the Gemini 2.0 Flash Thinking Experimental model, which enhances its Gemini app with advanced thinking and reasoning capabilities.

Google is dedicated to ensuring the safety and security of its AI models, employing automated red teaming and reinforcement learning techniques to mitigate risks.

The new model has shown improved performance across various benchmarks, including general capabilities and code generation tasks.

Google aims to leverage its extensive ecosystem of applications to create a superior AI agent that enhances user engagement and functionality.

The launch of Gemini 2.0 was touted as the release of the world's best AI model, according to AI expert @lmarena_ai.

The Pro version of Gemini 2.0 can handle a context of 2 million tokens, allowing it to process 1.5 million words in a single session and execute code while searching the web.

In terms of cost, processing data with Gemini Flash-Lite is priced at $0.019 per 1 million tokens, making it competitive against OpenAI's $0.075 and DeepSeek's $0.014, although DeepSeek plans to raise its prices significantly soon.

During the Q4 2024 earnings call, Google CEO Sundar Pichai downplayed the competitive threat posed by DeepSeek, highlighting the potential for cost-efficient AI models.

The reasoning capabilities of AI models like Gemini work by breaking down complex problems into smaller, manageable steps, which leads to more accurate solutions, albeit at a slower pace.

This strategy of lightweighting models reflects Google's commitment to enhancing AI performance while maintaining competitive pricing in the market.

This announcement is part of a series of rapid AI developments from Google, which also includes new features and enhancements across its platforms.