Google has launched its latest AI model, Gemini 2.5 Pro, touting it as its most advanced yet. Released in March 2025, Gemini 2.5 Pro is a reasoning model, meaning it is designed to “think” about and fact-check its response before generating them. Google says the model surpasses many of the industry’s top players, especially in coding, math and science benchmarks.
What Is Gemini 2.5 Pro?
Gemini 2.5 Pro is an AI model developed by Google that has native multimodal and advanced reasoning capabilities. Also known as Gemini 2.5 Pro Experimental, it is the inaugural model in the company’s larger Gemini 2.5 series.
Gemini 2.5 Pro is hitting the market at a particularly intense moment in the AI arms race, with companies like OpenAI, Anthropic, xAI, DeepSeek and Baidu all vying to develop models with increasingly sophisticated reasoning capabilities. In fact, this is Google’s second reasoning-focused release in just a few weeks, following closely behind Gemini 2.0 Flash Thinking. The move is a major precursor to its larger goal of developing agentic AI systems that can not only think like humans, but complete tasks like them too — an ambition shared by many of its peers.
“We’ve achieved a new level of performance,” Google said in a blog post. “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.”
Gemini 2.5 Pro is now available in Google AI Studio and in the Gemini app. Google also says the model will be available on its AI development platform, Vertex AI, “soon.”
What Is Gemini 2.5 Pro?
Gemini 2.5 Pro (also called Gemini 2.5 Pro Experimental) is the inaugural model in Google’s Gemini 2.5 series. Featuring native multimodal capabilities, Gemini 2.5 Pro can both process and interpret text, image, audio and video data within a single framework. It is also designed for advanced reasoning, where it methodically breaks down tasks and gradually builds a response instead of jumping straight to an answer like a typical language model. As such, the model is able to “analyze information, draw logical conclusions, incorporate context and nuance and make informed decisions,” according to Google, making it especially strong in areas like coding, math, science and logic.
What Can Gemini 2.5 Pro Do?
Google says Gemini 2.5 Pro has a wide range of advanced capabilities, including:
- Native Multimodality: The model can understand and integrate various different data types — namely text, images, audio and video — simultaneously, then provide text responses. For example, it can analyze a video and generate a relevant text summary.
- Coding Assistance: The model can be used to build all kinds of products, including “visually compelling” web applications and video games — and all from single line prompts.
- Agentic Coding: With its advanced reasoning capabilities, the model can independently write, modify, debug and refine code with little to no human supervision.
- Tool Utilization: Gemini 2.5 Pro can interact with other tools and services, allowing it to run external functions, execute code, structure data in specific formats (like JSON) and even perform searches. This ability helps the model complete tasks that involve multiple steps, communicate with other software through APIs and format its responses to fit specific systems.
- Mathematical and Scientific Problem-Solving: The model is especially good at tasks related to science and mathematics, demonstrating an ability to tackle complex, analytical problems.
- Large Context Window: With a context window of up to 2 million, the model is designed to process vast amounts of data, enabling it to analyze lengthy documents, handle extensive codebases and synthesize information from multiple sources within a single query.
How Does Gemini 2.5 Pro Compare to Other Models?
Google tested Gemini 2.5 Pro against other models of similar sizes and capabilities, including OpenAI’s o3-mini and GPT-4.5, Anthropic’s Claude 3.7 Sonnet, xAI’s Grok 3 and DeepSeek’s R1. These comparisons focused on areas like scientific reasoning, mathematics, multimodal problem-solving, coding and long-context tasks. Overall, Gemini 2.5 Pro outperformed (or came in close second to) most of its competitors.
Knowledge and Reasoning
One of Gemini 2.5 Pro’s most significant achievements, according to Google, was its “state-of-the-art” performance on the Humanity’s Last Exam evaluation, a dataset created by hundreds of different subject matter experts to test the breadth of a model’s knowledge and reasoning skills. With 18.8 percent of the answers correct , Gemini 2.5 Pro appeared to outpace its competitors’ scores here by a considerable margin.
Long Contexts and Multimodal Tasks
Gemini 2.5 Pro scored 91.5 in the long-context reading comprehension test, miles ahead of both OpenAI models (the only other ones tested on this). It also led the pack in multimodal understanding.
Math and Science
Gemini 2.5 Pro’s reasoning capabilities came through in the tests related to science and math, scoring high on AI benchmarks like GPQA Diamond, AIME 2024 and AIME 2025. However, Claude 3.7 Sonnet and Grok 3 performed marginally better on a couple of these tests.
Coding
On benchmarks that tested de-bugging, multi-file reasoning and agentic coding, Gemini 2.5 Pro performed well, but it did not dominate — trailing behind Claude 3.7 Sonnet and Grok 3. However, the model did come out ahead in code editing.
Gemini 2.5 Pro Limitations
Like any other language model, Gemini 2.5 Pro has the capacity to produce inaccurate responses, and it may reflect the biases present in its training data. It also did not perform as well in areas like code generation and factual accuracy, where OpenAI’s models were stronger.
How to Access Gemini 2.5 Pro
There are several ways to access Gemini 2.5 Pro, depending on your needs.
- Gemini App: This is probably the easiest way to try out Gemini 2.5 Pro on both mobile and web. It is now available for free, as well as to Gemini Advanced users.
- Google AI Studio: This will give you more control over inputs, tool use and multimodal prompts. It’s free (for now) and supports text, image, video, and audio inputs. After creating an account, just select Gemini 2.5 Pro from the model dropdown menu.
- Gemini 2.5 Pro API: For more flexibility, you can use the Gemini API to integrate the model into your own applications and workflows. You can call the model directly by enabling tool use, get structured responses and process long documents automatically. More technical details are available here.
Frequently Asked Questions
What can Gemini 2.5 Pro do?
Gemini 2.5 Pro can process and understand multiple types of data at once, including text, images, audio and video. It also excels at tasks that involve advanced reasoning, showing a particular strength in areas like math, science and coding.
Is Gemini 2.5 Pro free?
Yes, Gemini 2.5 Pro is available for free on the Gemini app.
Is Gemini 2.5 Pro available?
Yes, Gemini 2.5 Pro is available on the Gemini app and Google AI Studio. Google also says the model will be available on its AI development platform, Vertex AI, “soon.”
