Benchmarks, prices and stage coverage.

Every accuracy score, unit price and stage coverage check we publish lives on this page with its source and date. Where an alternative scores higher or runs faster, the table shows it.

What each vendor covers across the chain.

Columns compare CloudRaker with the five alternatives evaluated most often: Reducto, Extend, AWS Textract, Google Document AI and Azure Document Intelligence.

CloudRakerReductoExtendAWSGoogleAzure
Parse and OCRopen weights
Split and ClassifySeparate APIscustom splittercustom classifier
Cited ExtractForms, Tables, Queries
Audio Transcribebatch + WebSocketSeparate service (Transcribe)Separate serviceSeparate service
Decide (sub-second encoder)0.5–1.3 sFast classifier, no latency claim
Redact (documents and audio)Separate service (AI Language)
Fill and ComposeEdit only, no fill or composeFill only, no compose
Legally binding Sign0 creditsDetection only
Human approvals and MCPapprovals and MCPMCP only, no review queuereview queue and MCP

Document parsing on ParseBench.

rakedoc-nano-2 (1.2B, open weights) scores 80.1 overall, fifth on the leaderboard and first among open-weight models. Four closed managed tiers score higher overall, led by LlamaParse Agentic Plus at 86.7.

ParseBench overall score, leaderboard, including our submitted run
  1. LlamaParse Agentic Plus 86.7
  2. LlamaParse Agentic 82.8
  3. Pulse Ultra 2 81.6
  4. Upstage Document Parse 2 80.6
  5. rakedoc-nano-2 ours, open weights. $10 per 1,000 pages, $5.20 at enterprise volume 80.1
  6. anyformat 80.0
  7. KDL-Frontier-Parser-nano-v2 79.1
  8. Opus 5.5 78.8
  9. LlamaParse Cost Effective 77.7
  10. GPT-5.6 Sol 74.4
  11. Gemini 3 Flash 74.1
  12. Reducto (Agentic) $47.60 per 1,000 pages 70.9
  13. Azure Document Intelligence $10 per 1,000 pages 56.8
  14. Google Cloud Document AI $10 per 1,000 pages 48.7
  15. AWS Textract $15 per 1,000 pages 48.3

Published 9 November 2026.

Audio transcription speed and error rate.

Measured on a 10-minute two-speaker recording against vendor pricing pages and the Artificial Analysis speech-to-text leaderboard. rakeaudio-asr runs at 270x realtime (150x diarized) at $5.00 per 1,000 minutes with ~5% internal English WER; Deepgram Nova-3 is faster at 300x and Gemini 3.5 Transcribe is more accurate at 2.6% WER.

Speed on a 10-minute file, as a multiple of realtime
  1. Deepgram Nova-3 300
  2. rakeaudio-asr ours, transcription 270
  3. rakeaudio-asr, diarized ours, speakers separated 150
  4. Gemini 3.5 Transcribe 87
  5. OpenAI gpt-transcribe 42
  6. OpenAI gpt-4o-mini-transcribe 41
  7. OpenAI gpt-4o-transcribe 37
  8. OpenAI whisper-1 29

Side-by-side vendor teardowns.

Benchmark it on your own files.

Run 1,500 scanned pages or 3,000 audio minutes on the free tier, or pull rakedoc-nano-2 on your own GPU under AGPL-3.0.