ExtractBench v1.1: tighter boxes, new box ground truth, value fixes

README.md

@@ -66,6 +66,17 @@ ExtractBench contains 370 documents (4,869 pages) across 8 business domains and
 
 The [evaluation framework](https://github.com/run-llama/ExtractBench) runs pipelines and reports scores by split, tag, and system.
 
+## Updates
+
+- **v1.1 (2026-09-29)**
+  - Tighter bounding boxes for human-annotated documents
+  - New box ground truth for 103 real documents (grounding now covers 339 of 370)
+  - Minor value fixes
+  - Documents and schemas are unchanged
+- **v1.0 (2026-08-19):** Initial release: 370 documents, 4,869 pages.
+
+To load a specific version, pass `revision`, e.g. `load_dataset("llamaindex/ExtractBench", split="short", revision="v1.0")`.
+
 ## Dataset Introduction
 
 A document type is a family of documents that carry the same kind of information, such as SEC 13F filings, utility bills, or mortgage closing disclosures, however much their layouts differ. In ExtractBench, each document type has exactly one schema, shared by all of its documents.

long.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac5b22bbf810548a5e086e792d4afe0aa98dc572d24deceab820a2558570f3bd
-size 159965789
+oid sha256:b932c32a59a87a3e430cb475d2db94581aaf6a5589219579d2b43171cfe13b72
+size 163097453

medium.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ae0884d8e8dddb1282eb98282d942b65c8d99b82dd37e650edc53cd0f5bb5df
-size 62894509
+oid sha256:3fe712510db5cfaa80ff4469349ddc0071c70722dab26ed021bee218dac75b87
+size 64175355

short.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4b07b3612f1d4bcf4157d5b97b8692d671aeddaf96e5b5898338345b654e740
-size 29881640
+oid sha256:3373bfdd44077f852b8f60e1d5b878a582c0c9517b22bcc5f4ec0d28f2ada36c
+size 30717761