{"slug": "claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend", "title": "Claude Opus 5 and GPT 5.5 are surprisingly cheap for document parsing, but Extend is the only one that didn't hallucinate.", "summary": "A 119-call benchmark across seven document-parsing providers and 11 real-world documents found Extend was the only API with zero hallucinations and a perfect 1.00 table row F1 score, while Claude direct led raw field accuracy at 0.991 and GPT direct was cheapest at $0.00047 per correct field. AWS Textract failed twice, cost about 14x GPT, and took 45 minutes to set up, while Mistral and LlamaExtract each hallucinated 3 times. The test used Claude Opus 5 via /v1/messages and GPT-5.5 via /v1/responses, with Mistral OCR the fastest at a 4.2s median and Claude direct at 6.4s.", "body_md": "# Claude Opus 5 and GPT 5.5 are surprisingly cheap for document parsing, but Extend is the only one that didn't hallucinate.\n\nMost \"best of\" lists for document parsing are just people reciting pricing pages. To see what actually happens when you hit the API, I ran 119 calls across seven different providers using 11 real-world documents—everything from French bank statements and handwritten cheques to medical EOBs and photographed receipts. I used specific JSON schemas for each document type to see who could actually extract data without making things up.\n\n## Which API actually handles raw data and tables best?\n\nI looked at raw field accuracy and row F1 scores to see who missed the most data.\n\n- **Raw field accuracy:**[Claude](https://promptcube3.com/en/tags/claude/) direct won at 0.991. GPT direct and Reducto followed at 0.982.\n- **Hallucinations:** Extend had 0 hallucinations across 11 docs. Claude, GPT, and Reducto also hit 0. However, Textract failed twice, while Mistral and LlamaExtract both hallucinated 3 times.\n- **Table row recall:** Extend was the only provider with a perfect 1.00 row F1 score. Everyone else sat at 0.99.\n\nIf you're piping this data into another piece of software, a 0.99 accuracy score is a liability. A single hallucinated value can break a downstream process. In that sense, Extend is the safest bet for automated systems, even if Claude is slightly more accurate on raw fields.\n\n## How do the costs and speeds actually compare?\n\nThe price gap is wider than the marketing makes it seem.\n\n- **Cost per correct field:** GPT direct was the cheapest at $0.00047, with Mistral OCR close behind at $0.00049. AWS Textract was an order of magnitude more expensive—about 14x the cost of GPT.\n- **Speed:** Mistral OCR was the fastest with a median of 4.2s. Claude direct took 6.4s.\n\n## Which one is the biggest headache to implement?\n\nDeveloper experience varies wildly depending on whether the tool has a native schema.\n\n- **Fastest setup:** Claude and GPT took about 10 minutes to get running.\n- **Middle ground:** Extend took about 25 minutes.\n- **Slowest:** AWS Textract took 45 minutes and required an additional LLM step to normalize the data.\n\n[MCP](https://promptcube3.com/en/tags/mcp/), plugin, agents.md, and llms.txt. Reducto has 3 of those 4, while AWS has nothing specific for Textract.\n\n## The technical setup for the test\n\nI used the following endpoints and versions for this run:\n\n- **Extend:**`POST /extract` (API`2026-02-09` ,`extend-ai` 1.19.0)\n- **Reducto:**`POST /extract` (standard mode,`deep_extract` off)\n- **LlamaExtract:**`extract_stateless()` using`BALANCED` mode\n- **AWS Textract:**`AnalyzeDocument` (FORMS + TABLES) in us-east-1, paired with an LLM normalizer\n- **Mistral OCR:**`/v1/ocr` with`document_annotation_format` using`mistral-ocr-latest`\n- **Claude direct:**`/v1/messages` with forced tool using`claude-opus-5`\n- **GPT direct:**`/v1/responses` with`json_schema` using`gpt-5.5` (strict: false)\n\n[Next Jev clones are popping up everywhere just two days after launch →](https://promptcube3.com/en/threads/9523/)\n\n## All Replies （0）\n\nNo replies yet — be the first!", "url": "https://wpnews.pro/news/claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend", "canonical_source": "https://promptcube3.com/en/threads/9551/", "published_at": "2026-09-21 17:49:00+00:00", "updated_at": "2026-09-21 17:53:38.635240+00:00", "lang": "en", "topics": ["ai-tools", "ai-products", "large-language-models", "ai-agents", "agent-protocols"], "entities": ["Extend", "Claude Opus 5", "GPT-5.5", "Reducto", "LlamaExtract", "AWS Textract", "Mistral OCR", "Anthropic"], "alternates": {"html": "https://wpnews.pro/news/claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend", "markdown": "https://wpnews.pro/news/claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend.md", "text": "https://wpnews.pro/news/claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend.txt", "jsonld": "https://wpnews.pro/news/claude-opus-5-and-gpt-5-5-are-surprisingly-cheap-for-document-parsing-but-extend.jsonld"}}