Google introduces a new line of Gemini AI models
6 February 10:02
Google has announced a large-scale update of its flagship artificial intelligence system, introducing several new versions of Gemini 2.0. Each model has its own unique features and purpose, Komersant ukrainskyi reports.
Gemini 2.0 Flash: a new update for general use
Gemini 2.0 Flash, first presented at I/O 2024, has become popular among developers as a powerful working model. It is optimized for tasks with a large amount of data and a high frequency of requests.
The model can handle a variety of data types thanks to its multimodal approach and has an impressive context window of 1 million tokens. This allows it to analyze and understand huge amounts of information.
Soon, the model will receive additional capabilities, including image generation and text-to-speech. It is already available via APIs in Google AI Studio and Vertex AI, as well as in the Gemini app.
Gemini 2.0 Pro Experimental: the most powerful version for complex tasks
The new experimental version of Gemini 2.0 Pro is based on developer feedback and is Google’s most powerful model for programming and complex queries. It demonstrates the best results in understanding and analyzing world knowledge.
The key feature of the model is the largest contextual window of 2 million tokens. This allows it to comprehensively analyze and understand much larger amounts of information than previous versions.
The model also gained the ability to use additional tools such as Google Search and code execution. It is available to developers through Google AI Studio and Vertex AI, as well as to Gemini Advanced users.
Gemini 2.0 Flash-Lite: a cost-effective version for mass use
In response to the positive feedback on the price and speed of the 1.5 Flash model, Google introduced a new version 2.0 Flash-Lite. This model provides better quality than its predecessor, while maintaining the same speed and cost.
Like the full version of Flash, it has a context window of 1 million tokens and supports multimodal input. Google provides an impressive example of its effectiveness: the model can generate relevant captions for about 40,000 unique photos for less than a dollar in the paid version of Google AI Studio.
The new model is already available for public testing in Google AI Studio and Vertex AI, demonstrating better results than 1.5 Flash in most performance tests.
Google presented the following table with comparative characteristics of the models:
Security and liability
Google emphasizes its attention to security in the new Gemini 2.0 models. The entire lineup uses new reinforcement learning methods, where the model itself evaluates the quality of its answers.
The company has also introduced automated testing to assess security risks. Particular attention is paid to protection against indirect attacks through the introduction of malicious instructions into the data that can be processed by the AI system.
All new models support multimodal input with text output, and additional modalities will be available for general use in the coming months.
As wrote, Google’s AI will probably be able to be “censored” to deliver special content.
