Reader board
Which Python library reads PDF, Word, Excel and PowerPoint files without breaking? Which OCR engine reads images best, and which speech-to-text engine makes the fewest mistakes? Document libraries run on the same odd real-world files from the test collections of pdf.js, Apache POI and Apache Tika (bug reports, fuzzers, broken exporters); OCR and speech engines on images and audio with known text. Every new release on PyPI is tested again by itself.
Updated 06 November 2026. Lower is better in the 🔴 🟡 ⏱ columns; the table is sorted by them (the any-document table: by files turned into text). Click a library to
see every file it failed on, with the error, and what changed between its versions.
New releases as they are tested: Atom feed.
🏆 Best per format
PDFPyMuPDFfails on 6 of 724 readable filesWord (.docx)docx2txtfails on 1 of 103 readable filesExcel (.xlsx)python-calaminefails on 2 of 297 readable filesPowerPoint (.pptx)markitdownfails on 0 of 66 readable filesAny document → textKreuzbergtext from 374 of 477 filesOCR (image → text)PaddleOCRfinds 95.1% of the wordsSpeech to text (English)Parakeetword error rate 11.1%Language detectionpy3langid97.0% correct
1,077 files from Mozilla pdf.js test PDFs;
724 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | Text from (all files) | Errors (all files) | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|
| 1 PyMuPDF 2.1 | 6 | 7 | 0 | 711 | 14 | 2 | 13.2 |
| 2 pypdfium2 5.15.0 | 6 | 55 | 0 | 663 | 24 | 1 | 10.0 |
| 3 pypdf 7.0.0 since 6.21.4: ▼ 1 newly failing | 11 | 49 | 0 | 664 | 34 | 4 | 145.8 |
| 4 pdfminer.six 20260107 since 20251230: ▲ 2 fixed | 11 | 64 | 0 | 649 | 31 | 4 | 179.5 |
| 5 pdfplumber 0.11.10 since 0.11.9: ▲ 1 fixed | 11 | 64 | 0 | 649 | 32 | 5 | 227.6 |
| 6 markitdown 0.1.9 since 0.1.8: ▼ 1 newly failing | 12 | 64 | 1 | 648 | 31 | 94 | 516.3 |
Word (.docx)
130 files from Apache POI test files;
103 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | Text from (all files) | Errors (all files) | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|
| 1 docx2txt 0.9 | 1 | 0 | 0 | 102 | 16 | 1 | 2.9 |
| 2 mammoth 1.13.0 | 2 | 7 | 0 | 94 | 17 | 9 | 14.9 |
| 3 python-docx 1.2.0 | 2 | 13 | 0 | 88 | 17 | 3 | 4.3 |
| 4 markitdown 0.1.9 | 3 | 7 | 0 | 93 | 18 | 174 | 97.3 |
Excel (.xlsx)
353 files from Apache POI test files;
297 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | Text from (all files) | Errors (all files) | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|
| 1 python-calamine 0.9.1 | 2 | 2 | 0 | 293 | 26 | <1 | 2.0 |
| 2 openpyxl 3.1.5 | 5 | 6 | 0 | 286 | 28 | 6 | 19.4 |
| 3 markitdown 0.1.9 | 11 | 0 | 2 | 286 | 32 | 111 | 315.1 |
PowerPoint (.pptx)
95 files from Apache POI test files;
66 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | Text from (all files) | Errors (all files) | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|
| 1 markitdown 0.1.9 | 0 | 0 | 0 | 66 | 12 | 71 | 12.9 |
| 2 python-pptx 1.0.2 | 0 | 11 | 0 | 55 | 12 | 9 | 2.7 |
Any document → text
477 files from Apache Tika test documents;
406 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | Text from (all files) | Errors (all files) | Format not supported | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|---|
| 1 Kreuzberg 4.10.4 | 11 | 19 | 0 | 374 | 30 | 4 | <1 | 8.7 |
| 2 Unstructured 0.27.36 | 8 | 22 | 0 | 370 | 28 | 11 | 30 | 507.4 |
| 3 Docling 2.148.0 | 10 | 4 | 1 | 266 | 23 | 150 | 7 | 580.7 |
| 4 markitdown 0.1.9 | 4 | 17 | 1 | 253 | 18 | 149 | 125 | 388.2 |
By file type: files read with text
| Type | Kreuzberg | Unstructured | Docling | markitdown |
|---|---|---|---|---|
| .docx 71 files | 60/71 7 empty 4 failed | 57/71 10 empty 3 failed 1 not supported | 59/71 6 empty 6 failed | 57/71 10 empty 4 failed |
| .pdf 71 files | 45/71 22 empty 4 failed | 50/71 16 empty 5 failed | 52/71 14 empty 5 failed | 44/71 21 empty 6 failed |
| .doc 49 files | 37/49 5 empty 7 failed | 37/49 9 empty 1 failed 2 not supported | not supported | not supported |
| .rtf 44 files | 41/44 3 empty | 40/44 2 empty 2 failed | not supported | not supported |
| .html 38 files | 34/38 2 empty 2 failed | 34/38 2 empty 2 failed | 36/38 2 empty | 36/38 2 empty |
| .xlsx 35 files | 22/35 9 empty 4 failed | 21/35 10 empty 4 failed | 20/35 8 empty 7 failed | 20/35 9 empty 6 failed |
| .txt 30 files | 30/30 | 30/30 | 28/30 2 failed | 30/30 |
| .xls 30 files | 18/30 8 empty 4 failed | 18/30 7 empty 5 failed | not supported | 20/30 8 empty 2 failed |
| .pptx 29 files | 22/29 6 empty 1 failed | 21/29 7 empty 1 failed | 23/29 5 empty 1 failed | 22/29 6 empty 1 failed |
| .ppt 22 files | 16/22 6 empty | 16/22 4 empty 2 failed | not supported | not supported |
| .odt 17 files | 15/17 2 failed | 15/17 1 empty 1 failed | 15/17 2 failed | not supported |
| .msg 14 files | 14/14 | 14/14 | 13/14 1 failed | 14/14 |
| .eml 9 files | 9/9 | 9/9 | 9/9 | not supported |
| .odp 4 files | not supported | not supported | 2/4 2 empty | not supported |
| .ods 4 files | 3/4 1 empty | not supported | 4/4 | not supported |
| .csv 3 files | 3/3 | 3/3 | 3/3 | 3/3 |
| .epub 3 files | 3/3 | 3/3 | not supported | 3/3 |
| .json 2 files | 0/2 2 failed | 0/2 2 failed | not supported | 2/2 |
| .md 2 files | 2/2 | 2/2 | 2/2 | 2/2 |
OCR (image → text)
190 files from OCR test set (CORD receipts, DocLayNet pages, app screenshots);
190 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | ✅ Words found | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|
| 1 PaddleOCR 3.7.0 Note: With paddlepaddle 3.2.2, as its maintainers advise: with 3.3.1 (the default install) every image fails on CPU (PaddleOCR issue #18162). why | 0 | 0 | 0 | 95.1% | 3070 | 1,527.6 |
| 2 docTR 1.1.0 | 0 | 0 | 0 | 95.0% | 2286 | 864.8 |
| 3 RapidOCR 3.10.0 | 0 | 0 | 0 | 92.4% | 372 | 136.6 |
| 4 EasyOCR 1.7.2 | 0 | 0 | 0 | 75.2% | 3736 | 620.5 |
| 5 Tesseract 0.3.13 (tesseract 5.5.0) | 0 | 24 | 0 | 67.5% | 132 | 53.8 |
By kind of image: words found
| Images | PaddleOCR | docTR | RapidOCR | EasyOCR | Tesseract |
|---|---|---|---|---|---|
| documents 50 images | 91.6% | 92.0% | 90.2% | 71.5% | 88.2% |
| receipts 80 images | 94.8% | 95.5% | 89.4% | 67.7% | 37.6% |
| screenshots 60 images | 98.5% | 96.9% | 98.4% | 88.2% | 90.0% |
Speech to text (English)
130 files from Speech test set (LibriSpeech, FLEURS, AMI meetings, Earnings-22 calls);
130 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | ✅ Word error rate | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|
| 1 Parakeet 0.12.0 · model TDT 0.6B v2 Note: With huggingface-hub 1.31.0: with 1.32+ (new shared cache layout) and current onnxruntime, the model fails to load ("External data path escapes model directory"), as many other ONNX models do. why | 0 | 0 | 0 | 11.1% | 770 | 124.2 |
| 2 faster-whisper 1.2.1 · model base.en int8 Note: With PyAV 18.1.0: faster-whisper 1.2.1 (the latest release) fails on every file with PyAV 19 (fixed on its main branch, not released yet). why | 0 | 0 | 0 | 16.8% | 466 | 79.8 |
| 3 whisper.cpp 1.5.1 · model base.en | 0 | 0 | 0 | 16.8% | 1226 | 162.9 |
| 4 Whisper 20250625 · model base.en | 0 | 0 | 0 | 17.6% | 1142 | 164.9 |
| 5 Moonshine 20251121 · model base | 0 | 0 | 0 | 19.8% | 1228 | 167.9 |
| 6 Vosk 0.3.45 · model small en-us | 0 | 1 | 0 | 37.1% | 918 | 138.2 |
By kind of clip: word error rate
| Clips | Parakeet | faster-whisper | whisper.cpp | Whisper | Moonshine | Vosk |
|---|---|---|---|---|---|---|
| ami-close-mic 20 clips | 12.4% | 22.7% | 23.4% | 22.7% | 18.3% | 44.3% |
| ami-far-field 15 clips | 29.3% | 47.6% | 46.6% | 49.0% | 50.5% | 79.3% |
| earnings-calls 25 clips | 19.1% | 24.5% | 23.9% | 26.6% | 37.5% | 60.4% |
| fleurs 20 clips | 9.3% | 12.4% | 12.9% | 12.9% | 14.8% | 32.5% |
| librispeech-clean 30 clips | 1.3% | 3.6% | 3.3% | 3.6% | 3.6% | 10.3% |
| librispeech-other 20 clips | 6.4% | 10.9% | 11.8% | 11.5% | 10.6% | 23.8% |
Language detection
760 files from Language test set (Tatoeba sentences, 38 languages);
760 of them give text to at least one library.
| Library | 🔴 Fails where another reads | 🟡 Empty where another reads |
⏱ Hangs or crashes | ✅ Correct | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|
| 1 py3langid 0.4.0 | 0 | 0 | 0 | 97.0% | <1 | 0.5 |
| 2 Lingua 2.2.0 | 0 | 0 | 0 | 94.6% | <1 | 0.5 |
| 3 fast-langdetect 1.1.0 · model fastText lid.176 (lite) | 0 | 0 | 0 | 93.4% | <1 | 1.5 |
| 4 langdetect 1.0.9 | 0 | 0 | 0 | 93.0% | 1 | 1.3 |
| 5 langid 1.1.6 | 0 | 0 | 0 | 89.1% | 1 | 2.5 |
By kind of text: share correct
| Texts | py3langid | Lingua | fast-langdetect | langdetect | langid |
|---|---|---|---|---|---|
| paragraphs 190 texts | 99.0% | 97.4% | 95.8% | 99.5% | 99.5% |
| short-sentences 570 texts | 96.3% | 93.7% | 92.6% | 90.9% | 85.6% |
⚔️ Head to head
Any document → textmarkitdown vs DoclingAny document → textDocling vs UnstructuredAny document → textKreuzberg vs DoclingAny document → textmarkitdown vs UnstructuredAny document → textKreuzberg vs markitdownAny document → textKreuzberg vs UnstructuredPDFPyMuPDF vs pdfplumberPDFpypdf vs PyMuPDFPDFpdfplumber vs pypdfPDFpdfplumber vs pdfminer.sixPDFpypdf vs pdfminer.sixPDFPyMuPDF vs pypdfium2PDFmarkitdown vs PyMuPDFWord (.docx)python-docx vs docx2txtWord (.docx)mammoth vs python-docxWord (.docx)markitdown vs python-docxExcel (.xlsx)openpyxl vs python-calaminePowerPoint (.pptx)python-pptx vs markitdownOCR (image → text)EasyOCR vs TesseractOCR (image → text)PaddleOCR vs TesseractOCR (image → text)EasyOCR vs PaddleOCROCR (image → text)RapidOCR vs PaddleOCROCR (image → text)RapidOCR vs TesseractOCR (image → text)EasyOCR vs RapidOCROCR (image → text)docTR vs PaddleOCROCR (image → text)docTR vs TesseractSpeech to text (English)faster-whisper vs WhisperSpeech to text (English)faster-whisper vs whisper.cppSpeech to text (English)Vosk vs WhisperSpeech to text (English)faster-whisper vs VoskSpeech to text (English)Parakeet vs faster-whisperSpeech to text (English)Moonshine vs faster-whisperSpeech to text (English)whisper.cpp vs WhisperSpeech to text (English)Parakeet vs WhisperLanguage detectionlangdetect vs fast-langdetectLanguage detectionlangdetect vs LinguaLanguage detectionlangdetect vs langidLanguage detectionLingua vs fast-langdetectLanguage detectionpy3langid vs langidLanguage detectionLingua vs py3langid
How this is measured
- Each library runs a short recipe from its own docs (board/targets.py):
open the file, get all text. Installed with the extras the docs recommend for reading
(
pypdf[full]: AES decryption, Brotli), nothing else. One worker process per CPU; per file 60 s / 120 s / 180 s / 300 s at most (the any-document table allows longer: some tools load models), killed above the memory limit. - Fails where another reads: the library raised an error, hung or crashed on a file from which another library in the same table got text. Files that every library fails on (encrypted, truncated, fuzzed) are not counted against anyone.
- Empty where another reads: no error, but no text. This is often the recipe, not the library: for example
python-docx's paragraphs+tables recipe skips headers, text boxes and nested tables. Check the file list. - Any document → text: a file type the tool's docs don't list, or that the tool itself rejects as unsupported, counts as "not supported", not as a failure (lists in board/libs.json). Returning a file's raw RTF source counts as empty, for every tool. Each tool runs with default settings: no OCR switched on by hand.
- OCR: each image comes with the human-written ground-truth text published with its dataset. Words found is the share of its words that appear anywhere in the engine's output (reading order doesn't matter); an image that fails or comes back empty counts as 0. The ground truth of receipts and document pages covers only their annotated text, so a character error rate would punish engines for reading more than was annotated; it is in the raw results but not ranked. English models, CPU, default settings.
- Speech: each clip comes with its dataset's reference transcript. Both texts go through Whisper's English text normalizer (case, punctuation, "twenty" = "20"); word error rate is all substituted, missing and extra words over all reference words, the usual corpus-level WER. A clip that fails or comes back empty counts all its words as errors. The three Whisper implementations use the same base.en model; each other engine its documented default model. English, CPU, default decoding settings.
- Language detection: each text's language is known from Tatoeba. An answer counts as correct if the library's code matches after mapping variants to ISO 639-1 (zh-cn → zh, nb → no, iw → he). Close languages count as different: Croatian detected as Bosnian or Serbian, or Norwegian as Danish, is wrong. One call per text with default settings and all languages enabled.
- Text quality (order, tables, layout) is not ranked in the other tables; they are about not breaking. The raw results have every output's size and hash.
- Same files for every version: the test collections are pinned to one commit. GitHub-hosted runner, so times are comparable within one table, roughly.