Google With Major Update: Gemini 2.5

Gemini 2.5 Pro Thinking Leads The Benchmarks!

Google DeepMind sets new standards in the AI landscape with Gemini 2.5 Pro: The model leads the LMArena rankings by a clear margin and demonstrates impressive performance in critical areas.

The benchmarks speak for themselves: without additional costly techniques such as majority voting, Gemini 2.5 Pro achieves top scores on GPQA and AIME 2025, and with 18.8% on Humanity's Last Exam, it achieves state-of-the-art performance among models without tool support - a remarkable achievement on this challenging dataset compiled by hundreds of domain experts.

“Gemini 2.5 Pro is state-of-the-art across a range of benchmarks requiring advanced reasoning. Without test-time techniques that increase cost, like majority voting, 2.5 Pro leads in math and science benchmarks like GPQA and AIME 2025.” (Google)

Of particular relevance to the AI community, the massive context window of 1 million tokens (soon to be 2 million) enables the processing of large datasets while improving reasoning capabilities. This combination of expanded context and improved reasoning is particularly impressive in code performance, where Gemini 2.5 Pro achieves 63.8% on SWE-Bench Verified with a customized agent setup.

The seamless integration of multimodal capabilities completes the profile: text, audio, images, video and full code repositories can be processed simultaneously. Google has convinced across the board today!

Scaling and automation has a massive impact on global economic growth

OpenAI Gets Native Image Generation

OpenAI integrates native image generation capabilities into GPT-4o and Sora. The feature creates diagrams and infographics while allowing text-based image editing.