Google Launches Gemini 2.5 Pro. Better than Deepseek R1?

Google has launched Gemini 2.5, its latest AI model, featuring an experimental Pro version that “pauses and thinks” for more accurate and thoughtful responses.

Key Points

• Gemini 2.5 Pro (experimental) is available for developers and advanced subscribers, aiming to help user tackle increasingly complex problems.
• Gemini 2.5 Pro is available in Google AI Studio and the Gemini app (for $20/month Gemini Advanced subscribers), with Vertex AI support coming soon.
• Google’s Gemini 2.5 Pro surpasses previous models and competitors in key benchmarks: 68.6% in Aider Polyglot, 63.8% in SWE-bench (below Claude 3.7’s 70.3%), and 18.8% in Humanity’s Last Exam. It’s optimized for web apps and agentic coding.
• Gemini 2.5 Pro with a 1 million token capacity, it can process approximately 750,000 words per session and Google plans to expand support to 2 million tokens soon.
• Google also says Gemini 2.5 Pro can handle difficult problems from sources like audio, text, images, entire code repositories, and video.

Gemini 2.5 Pro: Advancements Over Gemini 1.5 and Industry Competitors

Google’s Gemini 2.5 Pro surpasses its predecessor, Gemini 1.5, with enhanced reasoning, a 1 million token context window, and improved accuracy. It Excels in multimodal tasks and contextual understanding.

It outperforms OpenAI’s GPT-4 Turbo in coding tasks like Aider Polyglot (68.6%) and SWE-bench Verified (63.8%), though Anthropic’s Claude 3.7 Sonnet leads with 70.3%. Results vary across benchmarks and are comparable across competitors.

Gemini 2.5 Pro also excels in complex problem-solving across text, images, code, and video.

Compared to DeepSeek R1, it demonstrates stronger performance in coding and reasoning, making it a competitive choice for developers and enterprises seeking advanced AI capabilities. However Deepseek v4 is expected to show comparable performance.

News Gist

Google has launched Gemini 2.5 Pro, an experimental AI model offering advanced reasoning with a 1 million token capacity.

It surpasses previous models in benchmarks, supports complex problem-solving across multiple formats, and is available for developers and Gemini Advanced subscribers.