LoRA earned its default status through a good accuracy-to-compute tradeoff, but the PEFT landscape has moved. This post puts several alternatives head-to-head across standard tasks and is one of the more useful practical comparisons published this cycle — the framing of "can you beat the most popular" forces a concrete answer rather than a survey. The guidance is particularly relevant if you are about to start a fine-tuning project and defaulted to LoRA without much deliberation. The post does not appear to declare a single winner, which is probably the honest answer.