What you need to know
- Google’s latest Gemini 2.0 experimental models are becoming widely available in the Gemini app for Android and iOS.
- Gemini 2.0 Flash Thinking Experimental, a thinking model, lets you trace Gemini’s thought process.
- The move is another step towards sunsetting Gemini 1.5 Flash and 1.5 Pro.
Google recently announced that it was reshuffling the AI models available in the Gemini app, sunsetting older Gemini models in favor of newer stable and experimental Gemini 2.0 models. Some of these changes are beginning to show up in the Gemini app for Android and iOS, with Gemini 2.0 Flash Thinking Experimental and Gemini 2.0 Pro Experimental rolling out today, Feb. 10.
The widespread rollout was first spotted by 9to5Google and was confirmed on Android Central’s devices. It makes the stable version of Gemini 2.0 Flash the default AI model for all users, and it’s described as being “for everyday tasks, plus more features.” This version of Gemini 2.0 Flash began to roll out at the end of January, replacing the experimental model.
Gemini 1.5 Pro and 1.5 Flash are still around, but are described as being the “previous model[s].” Google previously said Gemini 1.5 Flash and 1.5 Pro will stick around for “the next few weeks” in late January, so time is running out for these models. If you want to finish up existing conversation, now is a good time to do so. Replacing them are Gemini 2.0 Flash Thinking Experimental, 2.0 Flash Thinking Experimental with apps, and 2.0 Pro Experimental.
Gemini 2.0 Flash Thinking Experimental is “best for multi-step reasoning,” while 2.0 Flash Thinking Experimental with apps adds support for “reasoning across YouTube, Maps, and Search.” Gemini 2.0 Pro Experimental is “best for complex tasks,” and requires a Gemini Advanced subscription through Google One AI Premium.
The two thinking models are a first for Google, and they help users trace Gemini’s thought process as it navigates through a prompt. It first shows you a “Thoughts” section in real-time, which includes the various steps Gemini took to reach a conclusion. When it does, Gemini will provide a “Response” section with a final answer. This can be useful for understanding how AI chatbots reach solutions, and questioning whether further research is needed to confirm the accuracy of a Gemini response.
Google notes that as preview models, the Gemini 2.0 experimental models lack access to real-time information and some features.
These models, announced previously, are now widely available for use in the Gemini app. Soon, Google will clean up the list of Gemini models by sunsetting Gemini 1.5 Flash and 1.5 Pro.