Google’s Gemini 2.5 Pro AI Model Launched; Tops Leaderboard, Outperforms OpenAI’s o3 Mini

Google launched the successor to its Gemini 2.0 sequence synthetic intelligence (AI) fashions on Wednesday. Dubbed Gemini 2.5 Pro Experimental, it’s the first mannequin the corporate is releasing from the 2.5 household. The Mountain View-based tech big says that this sequence of fashions could have “thinking” or reasoning functionality constructed straight into the fashions. It additionally notes improved benchmark scores throughout a variety of capabilities, outperforming OpenAI’s o3-mini in a number of areas. Google has begun rolling out the mannequin to customers.

Gemini 2.5 Pro AI Model Released

In a weblog submit, Koray Kavukcuoglu, the CTO of Google DeepMind, detailed the brand new giant language mannequin (LLM). The most notable facet of the Gemini 2.5 sequence is that there’ll now not be any “Thinking” fashions such because the Gemini 2.0 Flash Thinking.

The tech big used an enhanced base mannequin, which was additional improved in post-training to ship inherent reasoning capabilities to all of the Gemini 2.5 AI fashions. Hence, Google won’t denote a selected “Thinking” label to a mannequin as all of them can perform superior reasoning and present chain-of-thought (CoT).

Gemini 2.5 Pro benchmarks
Photo Credit: Google

Google didn’t reveal rather a lot concerning the mannequin specs, so particulars round its dataset, coaching strategies, and structure aren’t identified. However, the tech big shared its benchmark scores primarily based on inside testing. It is alleged to have scored 18.8 % on Humanity’s Last Exam, a dataset thought of the hardest benchmarking take a look at for AI fashions. Gemini 2.5 Pro’s rating was state-of-the-art (SOTA) amongst fashions with out software use.

Gemini 2.5 Pro can be claimed to have outperformed fashions like OpenAI’s o3-mini, Grok 3 Beta, Claude 3.7 Sonnet, and DeepSeek R1 in a number of benchmarks, corresponding to GPQA Diamond, AIME 2024 and 2025, Aider Polyglot, and MMMU.

Apart from this, Gemini 2.5 Pro additionally ranked on the prime of the LMArena leaderboard at launch. LMArena is a user-based platform the place AI fans and builders fee fashions primarily based on their experiences. Currently, it’s adopted by Grok 3 preview, GPT 4.5 preview, Gemini 2.0 Flash Thinking, and Gemini 2.0 Pro for the second, third, fourth, and fifth positions, respectively.

Google claims that the newest LLM additionally improves coding efficiency and might create “visually compelling” internet apps and agentic code functions. Gemini 2.5 Pro additionally comes with native multimodal help and a context window of 1 million tokens.

Gemini 2.5 Pro is accessible to builders and enterprises by way of the Google AI Studio, and Gemini Advanced subscribers can entry the mannequin in Gemini’s internet consumer and apps. The firm plans to make it out there on Vertex AI within the coming weeks.