Grading Needs a Rubric, Not Intelligence
Small language models can grade open-ended examination answers as reliably as substantially more expensive models when they grade against an explicit rubric. We test this claim as the design principle behind any-to-bench: a frontier model reads source documents once, at ingestion, to extract each question and its rubric; lower-cost models then perform all repeated grading work. We evaluate six…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
