A LocalLLaMA community post argues that local models could fall further behind frontier models as hardware access and cost constrain what individuals can run. The poster says local models are currently months behind the latest frontier models.
The post’s author says local models have so far kept pace with frontier models with a lag of several months. They argue that access to hardware could widen the gap, citing the cost of hardware and frontier developers’ purchases for training.
The author expects model improvements to slow, but says hardware costs may not fall. These are the poster’s views; the post does not provide data or evidence on future model performance or hardware prices.