Google is rolling out four new experimental Gemini AI models, including ‘2.0 Flash Thinking’

Roughly a week after making Gemini 2.0 Flash available to all as a stable model, Google is now rolling out several new Gemini models — some designed for coding performance and complex prompts and the others for cost efficiency.

As highlighted by the tech giant in a new blog post today, one of its new models is a lighter version of the Gemini model you would normally rely on for daily tasks, and it comes in the form of Gemini 2.0 Flash-Lite.

Even though titled ‘Lite,’ Google says that the new model outperforms the regular Gemini 1.5 Flash model in a majority of benchmarks, only lagging behind when it comes to understanding long context and for code generation in Python.

It features the same 1 million token context window and multimodal input as the Gemini 2.0 Flash, albeit at the same speed and cost as 1.5 Flash. The new model is currently limited to Google AI Studio and Vertex AI in public preview.

On the other hand, new models like Gemini 2.0 Flash Thinking Experimental and 2.0 Flash Thinking Experimental with apps are available to try out within the Gemini app on both desktop and mobile.

Understand why Gemini said what it said

The new models are described as “best for multi-step reasoning” and for “reasoning across YouTube, Maps & Search,” respectively. The former is currently ranked as the world’s best model on the Chatbot Arena LLM Leaderboard, and is absolutely free to use. The primary advantage of using 2.0 Flash Thinking Experimental and 2.0 Flash Thinking Experimental with apps is the models’ ability to share their thought process. According to Google, “this model is trained to break down prompts into a series of steps to strengthen its reasoning capabilities…2.0 Flash Thinking Experimental shows its thought process so you can see why it responded in a certain way, what its assumptions were, and trace the model’s line of reasoning.”

Although not immediately beneficial at first, knowing the reasoning behind why the AI model said what it said can offer a greater level of transparency, allowing users to put more trust in the model’s output. Additionally, it can also help users understand the underlying logic behind the answer to a query, which can be a learning experience, as highlighted in the screenshots below.

Lastly, the tech giant also unveiled its current top-of-the-line Gemini 2.0 Pro Experimental, which it says is the “best model yet for coding performance and complex prompts.” Currently limited to Gemini Advanced users via the Gemini app on desktop and mobile, and to developers in Google AI Studio and Vertex AI, Gemini 2.0 Pro Experimental features an impressive 2 million token context window — which roughly equates to 3,000 pages full of text. The large window allows it to “comprehensively analyze and understand” information, integrate results from Google Search, and execute code.