v1.0.0

What's Changed

• Add Reducto agentic parsing configuration by @donald-reducto in #2
• Upgrade extend-ai SDK to 1.8.0 and enable richer parsing options by @ameya005 in #4
• Add Claude Opus 4.7 parse benchmark pipeline by @boyang-zhang1 in #5
• Add Qwen3.5-35B and Qwen3.6-35B FP8 parse benchmark pipelines by @boyang-zhang1 in #6
• Add GPT-5.4 Nano parse benchmark pipelines by @boyang-zhang1 in #7
• Add Gemini 3.1 Flash Lite layout pipelines and bump max_tokens by @boyang-zhang1 in #8
• Add Gemma 4 31B vLLM layout pipeline by @boyang-zhang1 in #9
• Add MinerU 2.5 vLLM parse pipeline by @boyang-zhang1 in #10
• Add leaderboard.csv with HF community sync by @boyang-zhang1 in #12
• Add Qwen3.5-9B / 2B / 0.8B vLLM layout pipelines by @boyang-zhang1 in #13
• Add Databricks ai_parse_document parse pipeline (single + batch) by @boyang-zhang1 in #15
• Add OpenAI GPT-5.5 parse-with-layout-file pipelines by @boyang-zhang1 in #16
• Add Pulse parse pipeline by @boyang-zhang1 in #18
• Fix --test flag silently ignored when full dataset already downloaded by @micahstubbs in #17
• Add regression test for --test auto-download path routing by @boyang-zhang1 in #19
• Add pulse_ultra_2 pipeline and fix Pulse layout adapter by @boyang-zhang1 in #22
• Add Granite Vision 4.1 4B pipeline and leaderboard entry by @boyang-zhang1 in #24
• Add full pulse_ultra_2 pipeline configuration and direct REST provider by @ritvikpandey21 in #25
• Add Docling Serve pipeline by @akreal in #21
• Add extract and parse-field grounding evaluation support by @SebasGarcia08 in #26
• Add local ParseBench annotator and visual grounding viewer apps by @SebasGarcia08 in #27
• Improve annotator sidebar resizing by @SebasGarcia08 in #28
• Serve visualizer favicon from static assets by @SebasGarcia08 in #29
• Register PaddleOCR-VL 1.5 and Falcon-OCR pipelines by @boyang-zhang1 in #32
• Add form_field parse rule for forms benchmark dimension by @boyang-zhang1 in #34
• feat: add infinity_parser2 by @ZumingHuang in #33
• Skip dataset auto-download when --input_dir is explicit by @boyang-zhang1 in #35
• Tolerate dash variants in page section rules by @LDD19 in #39
• Lazy-load anthropic and llama_cloud to avoid import-time dependency by @benjamin-sowell-glean in #37
• Add Gemini 3.5 Flash parse-with-layout pipelines by @boyang-zhang1 in #42
• Replace Extend beta pipeline with Extend Parse 2.0 by @boyang-zhang1 in #46
• Add Surya OCR 2 parse pipeline by @boyang-zhang1 in #45
• Add PaddleOCR-VL-1.6 parse pipelines by @boyang-zhang1 in #44
• Add Gemini 3.5 Flash pricing and leaderboard cost by @boyang-zhang1 in #47
• Add Anthropic Opus 4.8 parse-with-layout pipeline by @boyang-zhang1 in #43
• fix: blank pages were not being scored properly by @logan-markewich in #48
• Add Anthropic Fable 5 parse-with-layout pipeline by @boyang-zhang1 in #50
• Add MinerU 2.5 Pro 2605 pipeline + leaderboard entry by @boyang-zhang1 in #51
• Add KDL-Frontier-Parser-nano community leaderboard entry by @skihyeon in #49
• Fix databricks_ai_parse layout bbox normalization by @boyang-zhang1 in #52
• fix evals for zero results by @logan-markewich in #53
• add local-only OSS providers by @logan-markewich in #54
• Add LiteParse results by @logan-markewich in #55
• Add configurable LlamaParse API base URL by @Georgehe4 in #57
• Add Mistral OCR 4 parse provider and pipeline by @boyang-zhang1 in #58
• Add Unlimited-OCR parse provider and pipeline by @boyang-zhang1 in #59
• Add mistral_ocr_4_annotation (Document AI mode) by @boyang-zhang1 in #60
• Fix Datalab parse pricing and add image input support by @boyang-zhang1 in #63
• Add Anthropic Sonnet 5 parse pipeline by @boyang-zhang1 in #64
• Fast, exact-parity TEDS (Zhang-Shasha) + memoized GriTS LCS by @ajjimeno in #78
• Update Cost Effective leaderboard results by @boyang-zhang1 in #79
• Add GPT-5.6 sol/terra/luna parse pipelines and leaderboard rows by @boyang-zhang1 in #80
• Update Pulse runners to use public hosted API by @ritvikpandey21 in #65
• Add Warp Ingest parse pipeline by @JSv4 in #76
• Add MinerU-Diffusion and Nemotron-Omni parse pipelines by @boyang-zhang1 in #85
• Add Extend Light pipeline and refresh Extend leaderboard results by @ameya005 in #87
• Add oi_parser parse provider and pipeline by @thiagosantos1 in #81
• Keep formatting-rule matches within a single span by @dlaird-ant in #90
• Fix heading/formula delimiter doubling in items_to_markdown by @dlaird-ant in #89
• Add bbox_scale config for absolute-pixel layout coordinates by @dlaird-ant in #91
• Gate bbox_scale=None to providers that pin the image frame by @boyang-zhang1 in #92
• Add Gemini 3.6 Flash and 3.5 Flash Lite parse pipelines by @boyang-zhang1 in #93
• Improve PyMuPDF4LLM OCR pipeline and benchmark results by @hashiromer in #86
• Add Amazon Nova 2 Lite parse pipeline by @boyang-zhang1 in #105
• Canonicalize tables for Text Content evaluation by @LegendZZZZZ in #97
• Make LLM normalization off by default by @boyang-zhang1 in #107
• Add ExtractBench cross-link under the title in README by @boyang-zhang1 in #112
• Refresh LlamaParse leaderboard and add Agentic Plus by @boyang-zhang1 in #113
• Remove internal staging pipelines by @boyang-zhang1 in #114
• Add florin-parser-nano (open-weight VLM) by @ammoman21 in #99
• Add oi-parser to the leaderboard by @thiagosantos1 in #95
• Add Qwen and GLM ParseBench pipelines by @boyang-zhang1 in #119
• Add rakedoc-nano (open-weight VLM): provider, pipeline, docs, leaderboard entry by @baptistelaget in #120
• Add Anthropic Fable 5.1 parse pipeline and leaderboard entry by @boyang-zhang1 in #123
• Add reducto_r1 parse pipeline and leaderboard entry by @boyang-zhang1 in #124
• Update pulse_ultra_2 to preserve inline formatting by @morgan-pulse in #121
• 6. Package parse-bench for PyPI by @logan-markewich in #130
• Remove deprecated pulse pipeline and leaderboard row by @morgan-pulse in #122
• 5. Various parse scoring core fixes by @logan-markewich in #129
• Clarify Semantic Formatting scope in README by @boyang-zhang1 in #134
• 4. Port visual-grounding and extract scoring fixes by @logan-markewich in #128
• 3. Port provider, inference-runner and pipeline fixes; refresh liteparse by @logan-markewich in #127
• 2. Port evaluation-runner, aggregation and report fixes by @logan-markewich in #126
• 1. Add new rule types, metadata metrics and extension registries by @logan-markewich in #125
• vbump by @logan-markewich in #135
• Update CHANGELOG with versioning and remove section by @logan-markewich in #136

New Contributors

• @donald-reducto made their first contribution in #2
• @ameya005 made their first contribution in #4
• @boyang-zhang1 made their first contribution in #5
• @micahstubbs made their first contribution in #17
• @ritvikpandey21 made their first contribution in #25
• @akreal made their first contribution in #21
• @SebasGarcia08 made their first contribution in #26
• @ZumingHuang made their first contribution in #33
• @LDD19 made their first contribution in #39
• @benjamin-sowell-glean made their first contribution in #37
• @logan-markewich made their first contribution in #48
• @skihyeon made their first contribution in #49
• @Georgehe4 made their first contribution in #57
• @ajjimeno made their first contribution in #78
• @JSv4 made their first contribution in #76
• @thiagosantos1 made their first contribution in #81
• @dlaird-ant made their first contribution in #90
• @hashiromer made their first contribution in #86
• @LegendZZZZZ made their first contribution in #97
• @ammoman21 made their first contribution in #99
• @baptistelaget made their first contribution in #120
• @morgan-pulse made their first contribution in #121

Full Changelog: https://github.com/run-llama/ParseBench/commits/v1.0.0