Google DeepMind vice president Melvin Johnson discusses the Gemini release cycle and describes how the team approaches model development and deployment.
Johnson argues that the gap between closed frontier models and open-weight alternativesAn open-weight model makes its trained parameter values available so others can run or adapt it under the release's licence. has widened. This is his assessment, not an independently verified comparison, and he expects competition and rapid releases to continue.
The discussion also touches on product use, evaluation results and expectations for future models. The video is an interview with a Google executive and its claims should be read in that context.
Watch on YouTube
