Google makes the gemini “thought” model available on the application, launches Gemini 2.0 Pro

THE Google Gemini application Now includes a model of reasoning that shows its “reflection process”.

Wednesday, the technology giant announcement The availability of Gemini 2.0 Flash Thinking Experimental on the office and mobile versions of Gemini. In addition to doing its reasoning model, 2.0 Flash Thinking, more accessible to users, Google shared other announcements related to the Gemini family: the launch of Gemini 2.0 Pro, extended access to Gemini 2.0 Flash in AI Studio and Vertex AI, and a weak -comer called Flash -Lite.

Tweet may have been deleted

In Google, It looks like business as usual, because the company remains during the progression of its AI strategy to gradually add more capacities to its Gemini family. But the introduction of Gemini 2.0 Flash-Lite is an interesting timing given Deepseek disturbance Last week. In depth Fraked the AI ​​industry With its R1 model which is just as capable as competing models, but which has made a fraction of the cost, upsetting the beliefs that more money is equivalent to better models.

Google previously published a profitable model called 1.5 Flash, and CEO Sundar Pichai minimized The impact of Deepseek on Google in its call for results from the fourth quarter 2024 Tuesday, saying: “For us, it has always been obvious” that the models could become more profitable over time. That said, Google plans to spend $ 75 billion in capital spending, which is a huge increase of $ 32.3 billion in 2023.

Google’s announcements seem to cover all the basics of AI development: reasoning models, advanced models and low -cost models.

Announced in December, 2.0 Flash Thinking Rivals O1 from Openai And models of o3-mini reasoning in that it is capable of solving more complex problems and showing how it reasons them. Previously, he made his debut in AI Studio, but now has a wide availability in experimental mode as a drop -down option in the Gemini application.

Tweet may have been deleted

There is also a version that fits into YouTube, Google Maps and Google Search. With this version, the model can choose to search for answers on the web, draw relevant YouTube links or request Google Maps instead of only relying on its own training data.

For example, if you ask 2.0 a flash thought without integration of applications, “how long would it take to walk in China?” It would be based on its own knowledge base to reason through the vagueness and the various factors involved. But in the version with the aforementioned applications, he immediately goes to Google Maps to obtain answers (as indicated in his “thought”.)

The Gemini 2.0 Pro version is the most remarkable announcement in terms of technical progress. According to Google, 2.0 Pro is its most capable model, the best for coding and management of complex tasks (not to be confused with Flash 2.0 thought, which can also manage complex problems but shows its work.) Gemini 2.0 is available as an experimental version with advanced Gemini subscribers and in AI Studio and Vertex AI.

Gemini 2.0 Flash-Lite has a 1 million tokens context window and multimodal inputs and operates at the same speed and at the same price as the previous Flash version, so it has a punch for its size. Flash-Lite is available in preview mode on AI Studio and Vertex AI.