Google releases its most intelligent AI model with thinking built-in
What's the story
Google has unveiled the Gemini 2.5 Pro (experimental), its most advanced artificial intelligence (AI) model yet, for Advanced subscribers and developers. The launch marks another mid-year/model update in the Gemini series. The new model is part of Google's ongoing effort to enhance AI capabilities across all its models, enabling them to tackle more complex problems and support context-aware agents.
Enhanced capabilities
Gemini 2.5 series: A leap forward in AI reasoning
The Gemini 2.5 series, including the newly launched Pro model, marks a major leap in AI reasoning. These models are capable of thinking through their responses before replying, which leads to better performance and accuracy. Google has baked these thinking capabilities directly into all its models. The company also highlights Gemini's native multimodality as a key advantage, enabling it to interpret text, audio, images, video, and code.
Performance metrics
Gemini 2.5 Pro: A breakthrough in AI performance
The Gemini 2.5 Pro (codenamed "nebula") is the first model in this family, tailored for complex tasks. The model has scored a state-of-the-art 18.8% across models on Humanity's Last Exam, a dataset created by hundreds of subject matter experts to capture the human frontier of knowledge and reasoning. With this score, Gemini 2.5 Pro beats OpenAI's o3-mini, DeepSeek R1, and other rival AI models.
Coding advancements
Gemini 2.5 Pro: A game-changer for coding tasks
Gemini 2.5 Pro has made huge leaps in advanced coding, nailing the creation of visually stunning web apps and agentic code apps, as well as code transformation and editing. The model achieved an impressive score of 63.8% on SWE-Bench Verified, the industry standard for agentic code tests with a custom agent setup. This is a major improvement over its predecessor, the Gemini 2.0 model series. However, it is behind Claude 3.7 Sonnet which has a score of 70.3%.
Data processing
Gemini 2.5 Pro: A multimodal powerhouse
The Gemini 2.5 Pro model comes with a million-token context window (two million tokens coming soon), enabling it to understand massive datasets and solve complex issues from a variety of information sources such as text, audio, images, video, and even full code repositories. Google DeepMind CEO Demis Hassabis called Gemini 2.5 Pro "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."
Rollout
Gemini 2.5 Pro: Availability and future plans
The Gemini 2.5 Pro (experimental) is being rolled out to Gemini Advanced and Google AI Studio users first, with Vertex AI likely to follow suit in the coming weeks. Google plans to introduce pricing for the model soon, allowing users to access 2.5 Pro with higher rate limits for scaled production use. This rollout replaces last month's 2.0 Pro (experimental) in the Gemini app, giving access to apps (Gmail, YouTube, etc.) and file uploads.