Google has announced Gemini 2.5, its latest AI model that can handle complex reasoning and coding tasks. The release includes Gemini 2.5 Pro Experimental, which ranks first on the LMArena leaderboard and leads common coding, math, and science benchmarks.
1/ Gemini 2.5 is here, and it’s our most intelligent AI model ever.
Our first 2.5 model, Gemini 2.5 Pro Experimental is a state-of-the-art thinking model, leading in a wide range of benchmarks – with impressive improvements in enhanced reasoning and coding and now #1 on… pic.twitter.com/mtEdRCTcgF
— Sundar Pichai (@sundarpichai) March 25, 2025
“Gemini 2.5 models are thinking models, capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy,” said Koray Kavukcuoglu, CTO of Google DeepMind.
According to Google, the model’s reasoning capabilities extend beyond classification and prediction, allowing it to analyse information, draw logical conclusions, and incorporate context and nuance.
This new “thinking model” outperforms OpenAI o3 mini, GPT-4.5, DeepSeek-R1, Grok 3 and Claude 3.7 Sonnet in several benchmarks. It also achieves a state-of-the-art 18.8% among models without tool use on Humanity’s Last Exam, a dataset created by hundreds of subject matter experts to reflect the limits of human knowledge and reasoning.
“For a long time, we’ve explored ways of making AI smarter and more capable of reasoning through techniques like reinforcement learning and chain-of-thought prompting,” the company stated. “Now, with Gemini 2.5, we’ve achieved a new level of performance by combining a significantly enhanced base model with improved post-training.”
The model is available in Google AI Studio and the Gemini app for advanced users, with availability on Vertex AI expected soon. Google plans to introduce pricing in the coming weeks for higher-rate production use.
Developers and enterprises can start using Gemini 2.5 Pro in Google AI Studio now. “Going forward, we’re building these thinking capabilities directly into all of our models, so they can handle more complex problems and support even more capable, context-aware agents,” Google said.
“2.5 Pro excels at creating visually compelling web apps and agentic code applications, along with code transformation and editing,” the company said. On SWE-Bench Verified, Gemini 2.5 Pro scores 63.8% with a custom agent setup.
Google highlights improvements in Gemini 2.5’s context-handling capabilities. “2.5 Pro ships today with a 1 million token context window (2 million coming soon), with strong performance that improves over previous generations,” the company stated. The model is capable of comprehending text, audio, images, video, and full code repositories.
Gemini 2.5 follows the recent release of Google Gemma 3, the latest iteration in its Gemma family of open-weight models. It succeeds Gemma 2, which was released last year.
The tech giant also recently introduced Gemini’s native image generation in Gemini 2.0 Flash, which integrates multimodal input, advanced reasoning, and natural language processing (NLP) to produce high-quality visuals.
Google’s rival OpenAI has also launched image-generation capabilities in GPT-4o.
Meanwhile, DeepSeek on Monday announced a new update to its general-purpose AI model DeepSeek-V3. The updated model ‘DeepSeek V3-0324’ now ranks highest in benchmarks among all non-reasoning models.
Artificial Analysis, a platform that benchmarks AI models, stated, “This is the first time an open weights model is the leading non-reasoning model, marking a milestone for open source.” The model scored the highest points among all non-reasoning models on the platform’s ‘Intelligence Index’.
Moreover, recently, Reuters reported that DeepSeek plans to release R2 “as early as possible”. The company initially intended to launch it in early May but is now contemplating an earlier timeline.
