Benchmark
LongBench-Write
Check whether models can sustain useful writing at a requested length.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
120 prompts, evenly split between English and Chinese, with explicit length requirements and seven output types including creative, academic, news, and educational writing.
Methodology
A length-adherence score is reported separately from GPT-4o judgments of six quality dimensions. The combined score averages length and quality.