Gemini 2.0 Flash
Gemini 2.0 Flash, first introduced at Google I/O 2024, is now generally available through the Gemini API in Google AI Studio and Vertex AI. This model, designed for high-frequency and high-volume tasks, features a context window of 1 million tokens and supports multimodal reasoning. Developers can use it to build production applications across a wide range of domains. Future updates will include image generation and text-to-speech capabilities, Google said.
For regular users, 2.0 Flash has been integrated into the Gemini app on desktop and mobile platforms, allowing broader access to its capabilities for creating, interacting, and collaborating. Pricing details are available on the Google for Developers blog.
Experimental Release of Gemini 2.0 Pro
Google DeepMind has also released an experimental version of Gemini 2.0 Pro, which is optimised for coding and handling complex prompts. This model features the largest context window in the Gemini family at 2 million tokens, enabling it to process and analyse extensive information. It also includes tool-calling capabilities, such as Google Search and code execution.
Also read: OpenAI launches ‘Deep Research’: Here’s what it does and how to try it
The 2.0 Pro model is available to developers through Google AI Studio and Vertex AI, as well as to advanced users of the Gemini app. Its improved coding performance and reasoning abilities are expected to benefit developers working on sophisticated projects.
Gemini 2.0 Flash-Lite
A new cost-efficient model, Gemini 2.0 Flash-Lite, has been introduced in public preview. Flash-Lite offers improved quality over the 1.5 Flash model while maintaining the same speed and cost. It supports a 1 million token context window and multimodal input. For instance, it can generate captions for large sets of images at a low cost, making it suitable for applications requiring scalability and affordability.
Flash-Lite is now accessible in Google AI Studio and Vertex AI.
Multimodal Capabilities and Future Plans
All Gemini 2.0 models currently support multimodal input with text output, with additional modalities expected in future releases. These updates aim to enhance the versatility of the Gemini family for both developers and end-users.
Safety and Security Measures
Google DeepMind said the Gemini 2.0 models utilise reinforcement learning techniques to improve accuracy and handle sensitive prompts effectively. Automated red teaming has been implemented to identify and mitigate risks, including indirect prompt injection attacks, the company said.
Also read: Researchers flag link between DeepSeek chatbot and Chinese state telecom
