LongBench

LongBench

A bilingual (English + Chinese) benchmark for long-context understanding, with 21 tasks across document QA, summarization, few-shot learning, synthetic reasoning, and code completion. 4,750 test samples averaging 5k–15k tokens, scored with a fully automated evaluation.

MIT
1K – 10K
Text Generation
Text Classification
Question Answering
by @AIOZAI
2

Last updated: 15 days ago


Details
Files
Discussions
0

No discussions yet. Start the first one.

New Discussion