Weekly Update: Week of June 30, 2025

New AI model launches and records. Kicking off July, AI labs competed to showcase advancements. On July 1, Google’s DeepMind announced “Gemini 2”, an updated general AI system that reportedly surpassed OpenAI’s GPT-4 on several benchmarks, including coding and common-sense reasoning. Tech analysts noted that Gemini 2 is multimodal (processing text, images, and audio together) and could be a step toward more human-like reasoning. OpenAI, in response, teased their upcoming GPT-5 (as mentioned later) and in the interim released a fine-tuned model specialized in mathematics that clinched a gold medal score at the International Math Olympiad, stunning the academic world. This friendly rivalry highlights how AI innovation is accelerating through competition. Also this week, a coalition of researchers from universities (MIT, Berkeley) open-sourced a powerful new language model called Lion-3 that, while not as capable as commercial giants, can be used freely by anyone. This democratization through open-source ensures more people – including students and startups – can experiment and build with advanced AI, potentially spurring innovation at the edges. In sum, early July showed that the AI model race is yielding ever more capable systems at a rapid clip, each breakthrough quickly pushing others to follow suit, much to the benefit of overall AI progress.