Google has unveiled Gemini 2.5, the latest iteration of its artificial intelligence (AI) model family, designed to enhance reasoning capabilities. This new model family aims to set a new standard in AI performance by integrating advanced “thinking” processes that allow the AI to pause and analyze before generating responses. The company has stated that all future AI models will include these reasoning features.
Gemini 2.5 Pro Experimental
At the forefront of the new model family is Gemini 2.5 Pro Experimental, which Google claims is its most advanced model yet for tackling complex tasks. The model is multimodal, meaning it can process text, images, and other forms of data simultaneously. It is now available through Google AI Studio for developers and in the Gemini app for subscribers of the $20-a-month Gemini Advanced plan.
Read More: Google expands Gemini Live with AI-powered screen and video …
Performance Benchmarks
Google says Gemini 2.5 Pro outperforms previous models, as well as competing AI models from OpenAI, Anthropic, DeepSeek, and xAI. The model has demonstrated significant improvements in coding and reasoning tasks:
- Aider Polyglot (Code Editing Evaluation): Gemini 2.5 Pro scored 68.6%, beating OpenAI, Anthropic, and DeepSeek models.
- SWE-bench Verified (Software Development Abilities Test): It scored 63.8%, surpassing OpenAI’s o3-mini and DeepSeek’s R1, but falling behind Anthropic’s Claude 3.7 Sonnet (70.3%).
- Humanity’s Last Exam (Multimodal Knowledge Test): Gemini 2.5 Pro achieved an 18.8% score, outpacing most competing models.
The model also topped the LMArena leaderboard, a ranking system that measures AI performance based on human preferences.
Expanding Context Window for More Data Processing
One of the standout features of Gemini 2.5 Pro is its 1 million token context window, allowing it to process approximately 750,000 words at once—longer than the entire Lord of the Rings series. Google plans to expand this to 2 million tokens soon, making it one of the most memory-extensive AI models available.
A Leap Forward
With its enhanced reasoning capabilities, Gemini 2.5 Pro is set to revolutionize AI agents—autonomous systems that can perform tasks with minimal human intervention. The model’s ability to break down tasks into multiple steps and analyze data makes it particularly powerful for software development. Google researchers demonstrated its potential by prompting Gemini 2.5 Pro to create an endless-runner dinosaur game using HTML, CSS, and JavaScript in a single prompt, showcasing its coding proficiency.
TxGemma: AI for Drug and Therapy Development
Alongside Gemini 2.5, Google also introduced TxGemma, a suite of open AI models designed to enhance drug and therapy development. Built on Gemma 2, TxGemma uses large language models (LLMs) to predict drug properties, identify promising candidates, and forecast clinical trial outcomes.
TxGemma models come in three sizes—2 billion, 9 billion, and 27 billion parameters—and include specialized “predict” versions for narrow tasks such as drug classification and regression. The 9B and 27B versions also feature “chat” models, enabling researchers to ask questions and receive detailed reasoning.
Additionally, Google introduced Agentic-Tx, an AI agent powered by Gemini 2.0 Pro that integrates 18 tools for multi-step reasoning and molecular analysis. Researchers can use Agentic-Tx to streamline therapeutic research and drug discovery.
Read More: Meta sells over 1 million Ray-Ban smart glasses in 2024
Google is positioning Gemini 2.5 as a crucial step toward more capable AI systems that can perform complex tasks across industries. While pricing details for Gemini 2.5 Pro’s API remain undisclosed, the company plans to reveal them in the coming weeks.