Google's Gemini 2.5: Next-Gen AI with Enhanced Reasoning
Google has just launched Gemini 2.5, a groundbreaking family of AI reasoning models designed to "think" before responding. Leading the charge is Gemini 2.5 Pro Experimental, a multimodal AI model touted as Google's most intelligent creation yet.
Availability and Access
Gemini 2.5 Pro Experimental is now accessible through Google AI Studio, the company’s developer platform, and the Gemini app for Gemini Advanced subscribers (a $20/month plan). This move signals Google's commitment to integrating advanced reasoning capabilities into all future AI models.
The AI Reasoning Race
Since OpenAI's introduction of the first AI reasoning model, o1, in September 2024, the tech world has been in a frenzy to develop comparable or superior models. Anthropic, DeepSeek, Google, and xAI are now key players, all investing in AI reasoning models that leverage increased computing power and time for enhanced fact-checking and problem-solving.
The Power of Reasoning in AI
AI reasoning techniques are pushing the boundaries of what's possible in math and coding. Many experts believe these models are crucial for creating AI agents – autonomous systems capable of performing tasks with minimal human oversight. However, these advanced capabilities come at a higher cost.
Gemini 2.5 Pro: Performance and Benchmarks
Google's previous experiments with AI reasoning, including a "thinking" version of Gemini in December, paved the way for Gemini 2.5. This latest iteration represents Google's most ambitious attempt to surpass OpenAI's o series. Google claims that Gemini 2.5 Pro excels in creating visually appealing web apps and agentic coding applications, outperforming previous Google models and leading competitors on various benchmarks.
On the Aider Polyglot code editing evaluation, Gemini 2.5 Pro scored 68.6%, surpassing top AI models from OpenAI, Anthropic, and DeepSeek. However, on the SWE-bench Verified software development test, it achieved 63.8%, outperforming OpenAI’s o3-mini and DeepSeek’s R1, but falling short of Anthropic’s Claude 3.7 Sonnet (70.3%).
In the Humanity’s Last Exam, a multimodal test with thousands of crowdsourced questions from mathematics, humanities, and natural sciences, Gemini 2.5 Pro scored 18.8%, surpassing most rival flagship models.
Context Window and Future Plans
Gemini 2.5 Pro launches with a 1 million token context window, allowing the AI to process approximately 750,000 words in a single input – equivalent to the entire "Lord of The Rings" series. Google plans to soon double this to 2 million tokens.
Pricing details for the Gemini 2.5 Pro API are forthcoming, with Google promising to share more information in the weeks ahead.
Source: TechCrunch