With the new AI Gemini 2.0, Google finally responds to Deepseek and GPT-O3 Mini

The beginning of the year was particularly responsible for announcements around artificial intelligence. Between surprise Deepseek and its R1 model, Alibaba's response with Qwen 2.5,, Openai accusations against its Chinese competitors,, Sonar Perplexity,, Meta and his war rooms or the provision of O3-Mini andOperator By (again and again) OPENAI, the month of January (and us with it) was rolled out under the new features linked to artificial intelligence. However, a leading actor had remained remarkably discreet so far. It is of course Google, which seemed to mark the step in the new race for AI which “reason”. But it could obviously not last.
And it is therefore with a very slight late time on its main competitors that the online research giant has drawn three new models a few hours ago for its (now) Gemini 2.0 range. These are Gemini 2.0 Pro, for the treatment of complex tasks, Gemini 2.0 Flash Thinking, for reasoning, and Gemini 2.0 Flash-Lite, which relies on speed and energy efficiency.

Fierce competition

These additions complete the offer launched in December 2024 With Gemini 2.0 Flasha model that was already running after Openai and its GPT-O1. Since then, as said in the introduction, the environment of world AI has become more fierce than ever. The so-called reasoning models have multiplied, with Openai and Deepseek which are positioned on high-performance but accessible models, such as GPT-O3-Mini or Deepseek-R. Google has therefore adapted by offering several solutions, for different types of uses, ranging from complex processing of data to lighter and faster queries.

Gemini 2.0 Pro: Designed for complexity

Among the new features, Gemini 2.0 Pro arises as a jewel of the crown thanks to its extended treatment capacity. Available via Google AI Studio, Vertex AI and the Gemini application for subscribers to the paid offer, this model benefits from a contextual window of 2 million tokens, or about 1.5 million words. Concretely, it can theoretically treat the equivalent of the seven volumes of Harry Potter at once.
Designed for developers and users with substantial needs, it turns out to be effective for demanding tasks such as coding or solving complex problems, according to Google. He also incorporates external tools, including Google Search, and can execute live code, thus strengthening its analytical and understanding capacities.

Gemini 2.0 Flash Thinking and 2.0 Flash-Lite: reasoning or speed, of your choice

With Gemini 2.0 Flash Thinking, Google this time emphasizes the reasoning and transparency of the results generated. This free accessible model offers users the possibility of visualizing the AI ​​reflection stages.
Gemini 2.0 Flash-Lite is presented as a model that combines speed and low consumption of resources. He succeeds Gemini 1.5 Flash by offering improved performance without an increase in cost. It still has a very substantial contextual window of a million tokens and multimodal support (text, images).

Right in the Google ecosystem

These new models are not limited to autonomous use. Google gradually integrates them into its services such as YouTube, Maps and Search. In addition, they are accessible to developers via Google AI Studio and Vertex AI, in order to allow wider adoption, with for example integration into third -party applications. Google undoubtedly hopes to make these models omnipresent in the face of the solutions of Deepseek and Openai, which on their side on flexible and aggressive economic models. The Chinese company has notoriously struck very hard on this last point.

And the rest?

Despite these developments, certain points remain unanswered. Gemini 2.0 Flash Thinking, for example, is still in the test phase for certain features, including the generation of images and the vocal synthesis. Google also ensures that it will soon expand the multimodal capabilities of Gemini, while strengthening the safety and reliability of models via advanced learning methods.
Nevertheless, even if he should not undergo the same kind of pressure as Deepseek, the web giant has not yet communicated in detail on the technical specifications of these new models, which limits transparency around their design.