LlamaIndex releases Boo Bench, a 179-page parser stress test, and reports LlamaParse at 179/179

LlamaIndex releases Boo Bench, a 179-page benchmark built to push document parsers into repetition loops, with a leaderboard, paper, and dataset on its site and Hugging Face.

In one case the company says a model saw 246 redaction x's on a page and wrote 23,004 of them.

LlamaIndex says its LlamaParse Agentic and Agentic Plus modes scored 179 of 179 with zero failures on the benchmark. It reports the best frontier vision-language models, GPT-6 Luna and GPT-6.1 Terra, reached 99.4%, while open vision-language models scored between 61% and 93%. These figures are LlamaIndex's own results on its own benchmark.