Perplexity introduces JOLT, a post-training method using one model as teacher and student

Perplexity researchers introduce JOLT, a post-training method in which one model acts as both teacher and student, with the teacher trained to give feedback the student can use.

Perplexity says that on the MATH benchmark, JOLT matched the accuracy of GRPO while using 6.5x fewer training tokens. The claim comes from Perplexity's own announcement on X.