In a strategic move ahead of its annual developer conference, Google has introduced Gemini 2.5 Pro Preview, an enhanced version of its flagship AI model. This release underscores Google’s commitment to advancing artificial intelligence capabilities, particularly in coding and multimodal processing.
In This Article:
Enhanced Coding Capabilities
Google DeepMind CEO Demis Hassabis announced the launch of Gemini 2.5 Pro Preview, referring to it as “the best coding model we’ve ever built.” The model demonstrates significant improvements in various coding domains, including front-end development, code transformation, and the creation of interactive applications. It ranks No. 1 on LMArena in coding and leads the WebDev Arena Leaderboard, showcasing its prowess in building interactive web apps and aiding in rapid prototyping.
Multimodal Reasoning and Extended Context
Gemini 2.5 Pro is natively multimodal, capable of processing text, audio, images, video, and code. This inherent design allows the model to handle complex tasks involving diverse data types seamlessly. Moreover, it supports a context window of up to 1 million tokens, with plans to expand to 2 million, enabling it to process extensive datasets and maintain coherence over long interactions.
Benchmark Performance
The model has achieved top performance on several benchmarks, surpassing competitors like OpenAI, Anthropic, xAI, and DeepSeek. Notably, it scored 92.0% accuracy on the AIME 2024 dataset, designed to assess advanced algebra and number theory problem-solving skills. In coding benchmarks, Gemini 2.5 Pro scored 70.4% on LiveCodeBench and 74.0% on the Aider Polyglot benchmark, indicating strong performance in code generation and editing tasks.
Developer Access and Pricing
Initially available to Gemini Advanced subscribers, Google has now expanded access to Gemini 2.5 Pro, making it free for all users. Advanced subscribers still benefit from additional features like higher request limits and longer context windows. For developers, Google has introduced a two-tier pricing model: up to 200,000 tokens per month at $1.25 per million input tokens and $10 per million output tokens; beyond that, prices increase to $2.50 and $15, respectively.
Integration and Future Prospects
Gemini 2.5 Pro is integrated into various Google platforms, including AI Studio and the Gemini app, with plans to extend to Vertex AI. The model’s capabilities are also being leveraged in applications like NotebookLM, enhancing research functions and supporting multilingual users. As Google continues to innovate in the AI space, the launch of Gemini 2.5 Pro sets a new benchmark in artificial intelligence, offering developers and users advanced tools for a wide range of applications.
By – Sonali