Google released a new family of AI reasoning models on March 25 under the Gemini 2.5 range, beginning with the experimental Gemini 2.5 Pro.
The tech conglomerate added that Gemini 2.5 Pro Experimental ranked first on the LMArena leadership board, insisting on the model’s superiority as state-of-the-art technology across different debuts and benchmarks. The model notably ranked higher than xAI’s Grok 3 Preview and OpenAI’s GPT 4.5 Preview.
The company further highlighted that the Gemini 2.5 models are the most intelligent Google released and should handle more complex tasks. Google also announced that all its upcoming AI models will have reasoning as an integrated capability. The tech company added that the Gemini 2.5 family included thinking models that could reason before providing answers to user questions, making the models showcase higher accuracy and performance.
Google released Gemini 2.5 Pro Experimental to users on March 25 through Google AI Studio or the Gemini App for subscribers to the Gemini Advanced plan. The company also promised that the model would be coming to Vertex AI soon, but it did not provide a definitive date. The team further mentioned that the pricing model for the Gemini 2.5 Pro would also be available to enable Google AI users to leverage “2.5 Pro with higher rate limits for scaled production use.”
The Gemini 2.5 Pro release comes a few months after the company launched its first reasoning model, Gemini 2.0 Flash Thinking. The launch was a step toward providing viable competition for OpenAI’s o1. Google rolled out Gemini 2.0 Flash and Flash Lite widely at the end of February through Google AI Studio and Vertex AI for enterprise users. The DeepMind team explained that users could utilize the models for Voice AI, data analysis, image editing, and more.
Google focuses on enhanced reasoning for upcoming models
Gemini 2.5 Pro is an awesome state-of-the-art model, no.1 on LMArena by a whopping +39 ELO points, with significant improvements across the board in multimodal reasoning, coding & STEM. You can try it out now in AI Studio https://t.co/lLpF8ToTVJ & @GeminiApp with Gemini Advanced https://t.co/bgjabz8O1u
— Demis Hassabis (@demishassabis) March 25, 2025
Google DeepMind CTO Koray Kavukcuoglu explained in the Gemini 2.5 models’ announcement that the company had been heavily gravitating toward making AI smarter. Kavukcuoglu further mentioned that the company aimed to also boost the upcoming models’ reasoning using techniques like chain-of-thought prompting and reinforcement training. The DeepMind CTO added that reasoning in the models went beyond simple classification and prediction. Kavukcuoglu revealed that reasoning involved the AI’s ability to deduct and analyze information, settle on more logical conclusions, make more informed decisions, and apply nuance and context.
Google focused on improving the performance level in Gemini 2.5 by combining a superior base model with better post-training. The DeepMind CTO highlighted that the incorporation of the reasoning capabilities in later models would help them support more context-aware agents while handling complex tasks. The announcement also revealed that users could submit their honest feedback on the model to help Google improve its AI models.
OpenAI was notably the first tech company to introduce a reasoning AI model in September 2024, releasing the OpenAI o1. The tech company revealed in an announcement that the o1 model evaded reasoning pitfalls related to AI development. OpenAI explained that its generative AI models employed fact-checking by considering different sections of a question posed by users.
Tech companies dedicate more funds toward AI reasoning development
Big tech companies, including Microsoft Corp., OpenAI, Meta Platforms Inc., Amazon.com Inc., and Nvidia, are investing more in AI infrastructure and reasoning. The companies ranked as hyperscalers plan to contribute $371 billion toward computing resources and building data centers. This amount increased by 44% this year compared to last year and is expected to surge to $525 billion by 2032.
OpenAI and China’s DeepSeek are some of the companies focusing part of their investments on AI reasoning, raising competition in the AI development race. The LMArena Leaderboard ranked OpenAI’s GPT 4.5 Preview and o1 and DeepSeek’s DeepSeek R1 among the top 10 AI reasoning models. A recent report further highlighted that investing in reasoning models could raise software profits.
Cryptopolitan Academy: Coming Soon - A New Way to Earn Passive Income with DeFi in 2025. Learn More
