Cohere's Parse 5: RAG parsing model at $1.5 per 1k pages, explained by staff

Releases·Model Release

Cohere launches Parse 5 vision parsing model with ParseBench top score

Cohere released Parse 5, a vision parsing model that converts complex files like tables and embedded images into clean Markdown with bounding boxes. It scores 79.2 on ParseBench, outperforming Mistral's 74.5 and Azure Document Intelligence's 74.3. The model targets RAG systems, document indexing, and agentic retrieval.

Cohere released Parse 5, a vision parsing model that converts complex files like tables and embedded images into clean Markdown with bounding boxes. It scores 79.2 on ParseBench, outperforming Mistral's 74.5 and Azure Document Intelligence's 74.3. The model targets RAG systems, document indexing, and agentic retrieval.

A developer released Chunkr, an open-source Rust chunking library supporting strategies like recursive, markdown header, late, and hierarchical chunking, plus native PDF loading. Benchmarks posted on an M4 MacBook show throughputs up to 3,232 MB/s and PDF extraction 15.9x faster than PyPDF, though Chonkie wins on BPE tokenization.

Products·launched

Sygal: local-first document conversion to Markdown + JSON for RAG pipelines

Sygal converts PDFs, Word, Excel, HTML, email, EPUB and scans into clean Markdown and structured JSON for RAG pipelines, running locally so documents never leave the machine. LLMs are used for OCR of scans/images via user-supplied API keys to providers like OpenAI, Claude, Mistral, NVIDIA, or Gemini. It solves the problem of layout noise and cloud-privacy concerns when prepping documents for embedding.

Sygal converts PDFs, Word, Excel, HTML, email, EPUB and scans into clean Markdown and structured JSON for RAG pipelines, running locally so documents never leave the machine. LLMs are used for OCR of scans/images via user-supplied API keys to providers like OpenAI, Claude, Mistral, NVIDIA, or Gemini. It solves the problem of layout noise and cloud-privacy concerns when prepping documents for embedding.

A simple API that takes any webpage URL and returns clean Markdown. Useful for AI applications, RAG pipelines, search engines, and content automation workflows that need structured text from web pages.

Products·launched

Browser-based PDF to Markdown converter using heuristics, not LLMs

LiteDoc is a browser-based tool that converts PDFs to Markdown using layout heuristics and OCR, avoiding the cost and complexity of LLM-based APIs. It solves the problem of expensive multi-step pipelines for users who need clean Markdown from PDFs without running a local model or paying per page.

LiteDoc is a browser-based tool that converts PDFs to Markdown using layout heuristics and OCR, avoiding the cost and complexity of LLM-based APIs. It solves the problem of expensive multi-step pipelines for users who need clean Markdown from PDFs without running a local model or paying per page.