Summary

  • Google has launched its updated AI model, Gemini 2.5, which combines an improved base model with post-training enhancements to provide what the company claims is the best overall performance of any AI model currently on the market.
  • Gemini 2.5 is natively multimodal, enabling it to interpret not only text but also audio, images, video and code, and Google plans to soon increase the token context window from 1 million to 2 million to enable the model to process more data.
  • The company has released a video demonstrating how Gemini 2.5 can use reasoning capabilities to create a video game from a single prompt asking it to make “a game where a horse runs around a track and jumps over obstacles.”
  • Google’s CEO of DeepMind, Demis Hassabis, claims the new model is “an awesome state-of-the-art model,” adding that it has “significant improvements across the board in multimodal reasoning, coding and STEM.

By Richard Lawler

Original Article