Unveiling the Evolution of Artificial Intelligence
Have you ever paused to reflect on the automatic actions or words shared in daily interactions? Often, we act on impulse, without deeper consideration. Until recently, artificial intelligence mirrored this behavior. Now, with the advent of reasoning models, the landscape has shifted dramatically. Google has announced a groundbreaking development for its premier AI model: Gemini 2.5 Pro, featuring the innovative ‘Deep Think’ capability.
Introducing ‘Deep Think’ at Google I/O
Revealed during Google I/O, ‘Deep Think’ empowers AI to evaluate multiple outcomes before delivering answers, markedly enhancing its capacity for complex problem-solving and reasoning assessments. Although Google has withheld specific operational details, it is anticipated that the mechanism aligns with the solution synthesis engines found in OpenAI’s o1-pro and the forthcoming o3-pro models.
Leading the Charge in Coding and Multi-Skill Domains
Google reports that with Deep Think, Gemini 2.5 Pro ascended to the top of LiveCodeBench, a rigorous coding benchmark. It also surpassed OpenAI’s o3 in the MMMU evaluation, which assesses multimodal perception and reasoning capabilities. Currently, ‘Deep Think’ is accessible solely to ‘trusted testers’ and is undergoing a restricted preview phase via Google’s API platform. Google underscores the comprehensive security assessments conducted prior to expanding access.
Enhancements in Gemini 2.5 Flash and the Arrival of Gemini Diffusion
Google has also refined Gemini 2.5 Flash, its cost-efficient AI variant, now excelling in coding, multimodal analysis, extended text contexts, and logical reasoning. This version operates with greater efficiency than its predecessor. Developers can explore this upgraded model through Google’s AI Studio, Vertex AI, and Gemini applications, with full availability anticipated by June.
Additionally, Google introduced Gemini Diffusion, a model noted for generating outputs 4 to 5 times faster than similar-sized rivals and competing effectively with models twice its scale. For now, Gemini Diffusion is also confined to ‘trusted testers.’
New Text-to-Speech Features
Lastly, Google has launched new text-to-speech preview features for Gemini 2.5 Pro and Flash models. These features, which support 24 languages and offer two distinct voice options, are available through the Gemini API.
News
