NewsNTech
The structural constraint in the large language model race is the training cycle: when a rival publishes a stronger model, the trailing lab cannot catch up with a patch.
It must run a new training cycle, and that clock is the pressure Koray Kavukcuoglu now carries as the new head of Google DeepMind. His stated task is to keep Gemini competitive against OpenAI and Anthropic.
Where the race actually runs Frontier models compete on capability evaluations, benchmarked against one another in near real time.
Each new model release from OpenAI or Anthropic resets those rankings and advances the standard.
Keep reading