Google has launched a new reasoning model within its Gemini app that showcases its “thought process.”

On Wednesday, the tech corporation revealed the rollout of **Gemini 2.0 Flash Thinking Experimental** for both desktop and mobile platforms of the Gemini app. In conjunction with this update, Google also rolled out **Gemini 2.0 Pro**, broadened accessibility to **Gemini 2.0 Flash** in AI Studio and Vertex AI, and presented a **cost-effective model named Flash-Lite**.

### Google Broadens Its AI Portfolio

Google is persistently refining its AI abilities by progressively advancing the Gemini series. Nonetheless, the debut of **Gemini 2.0 Flash-Lite** occurs at a notable moment, following the disruption instigated by **DeepSeek** last week. DeepSeek’s **R1 model** astonished the AI sector by achieving performance on par with top models but at a considerably reduced cost, questioning the belief that larger budgets are always linked to superior AI.

Previously, Google had introduced a budget-friendly model, **1.5 Flash**, and CEO **Sundar Pichai** minimized DeepSeek’s influence during the company’s **Q4 2024 earnings call**. He remarked that it was “always clear” that AI models would evolve to be more cost-effective over time. Despite this, Google intends to **boost capital expenditures to $75 billion**, a notable increase from **$32.3 billion in 2023**.

### A Comprehensive AI Approach

Google’s recent announcements encompass every facet of AI innovation:

– **Reasoning models** (Gemini 2.0 Flash Thinking)
– **Advanced models** (Gemini 2.0 Pro)
– **Cost-efficient models** (Gemini 2.0 Flash-Lite)

Initially launched in December, **2.0 Flash Thinking** competes with **OpenAI’s o1 and o3-mini models** by tackling complex issues and clarifying its reasoning. Initially offered in AI Studio, it is now available in experimental mode as a **dropdown option in the Gemini app**.

### Integration with Google Services

A revamped version of **2.0 Flash Thinking** integrates with **YouTube, Google Maps, and Google Search**. This enables the model to browse the internet, supply pertinent YouTube links, or utilize Google Maps rather than depending solely on its training data.

For instance, if a user inquires, *”How long would it take to walk to China?”*, the standalone model would evaluate factors based on its internal understanding. In contrast, the integrated version would **rapidly consult Google Maps** to provide an answer, illustrating its reasoning approach.

### Gemini 2.0 Pro: The Pinnacle Model

The most noteworthy technical upgrade is **Gemini 2.0 Pro**, which Google asserts is its **most competent model**—perfect for **coding and managing intricate tasks**. Unlike **2.0 Flash Thinking**, which articulates its reasoning, **2.0 Pro prioritizes performance**. It is accessible as an **experimental version** for **Gemini Advanced subscribers**, as well as in **AI Studio and Vertex AI**.

### Flash-Lite: An Economical AI Model

The **Gemini 2.0 Flash-Lite** model boasts a **1 million token context window**, accommodates **multimodal input**, and operates at the same **speed and price** as the preceding **1.5 Flash** model. Despite its reduced size, it continues to be a formidable tool. **Flash-Lite is available in preview mode** on **AI Studio and Vertex AI**.

With these enhancements, Google is bolstering its AI ecosystem, ensuring it remains a competitor in the swiftly changing AI arena.