Google’s Gemini 1.5 Flash and Gemini 1.5 Pro models launched with support for a one million token context window and multimodal capabilities. Then came the Gemini 2.0 Flash and Gemini 2.0 Pro models, released experimentally with a more efficient design for developer applications.
Powerful as they were, one had limited availability, and the other still needed more optimization. Now, Google is making them widely available, as well as introducing the Gemini 2.0 Flash-Lite as a cost-efficient option.
Meet the new Gemini AI models
Google is making Gemini 2.0 Flash widely available through the Gemini Application Programming Interface (API) in Google AI Studio and Vertex AI. Developers can now use it for various applications.
They’re also introducing Gemini 2.0 Pro Experimental. It can handle coding and complex questions with a more significant context window of two million tokens. The model will be available to Gemini Advanced subscribers.
Gemini 2.0 Flash-Lite is another new model with a one million token context window. Google says it’s meant to be a cheaper version of Flash, but it is now in public preview. If you had 40,000 unique photos, the model can generate a caption for each one. All these different models can handle text, images, and more. But for now, we’ll only see text-based output, with the remaining features arriving later.
Related: Verizon’s New Deal Offers Gemini Advanced at Half the Price
Gemini 2.0 models pricing
Google has shared detailed pricing for the Gemini 2.0 family, and the Gemini 2.0 Flash model charges $0.10 per million tokens for text, image, or video inputs, while audio inputs are priced at $0.70 per million tokens. Text outputs are set at $0.40 per million tokens, while context caching for text, image, or video inputs will cost $0.025, and audio context caching is priced at $0.175.
For developers looking for a more affordable option, Gemini 2.0 Flash-Lite charges $0.075 for text, image, or video inputs and $0.075 for audio inputs. Text outputs cost $0.30 per million tokens, and context caching costs $0.01875.
For comparison, the Gemini 1.5 Flash model has different pricing based on the size of the prompts. Smaller prompts using less than 128,000 tokens have a $0.075 input cost, and text outputs are $0.30.
Larger prompts above 128,000 tokens cost $0.15 for inputs and $0.60 for outputs. Context caching varies as well, depending on the size of the prompts. Audio input pricing will change on February 20, 2025, and context caching for audio is coming soon.
Take advantage of Gemini 2.0’s free tier
Google offers a tier that lets you experiment without cost, and you can also scale as you go. When your app grows, and you need more processing power, you can increase usage to handle larger amounts of data or more frequent requests, but you’ll still be within the company’s rate limits.
They are also committed to using automated red teaming to test the safety of AI systems. In simulating attacks where malicious instructions hide in AI data, they can stop models from responding to harmful commands.
Although indirect prompt injection attacks aren’t as common as other types of cyberattacks, they are becoming more of a concern as AI systems increase in popularity. Since they don’t always differentiate between harmless and harmful information unless specifically structured to do so, it’s a necessary security measure.
