Resources

PageIndex Changelog

Nov 4, 2026

  • PageIndex Flash on Cloud. PageIndex Flash now runs on Cloud. It indexes text-based PDFs at $0.001 per page, one tenth of Cloud's default indexing rate ($0.01 per page).

Oct 27, 2026

  • Introducing Page Search API. Find the pages that answer a question in about a second, and send your LLM only those pages, not the whole document. It cuts token cost by 97%, and is more accurate and faster than vector search (benchmark).

Oct 23, 2026

  • PageIndex Financial QA Benchmark. Compare the accuracy and query cost of LLMs answering questions about annual reports (Form 10-K) with PageIndex. Frontier models tested include Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon, as well as Kimi K3.1 Pro, Qwen3.8 Max and DeepSeek V4.1 Flash.

Oct 19, 2026

  • Playground in Dashboard. Try different LLMs with PageIndex on your own documents, directly in your browser, and compare their answers and token costs. No code needed.

Oct 18, 2026

Oct 12, 2026

  • Cheaper than vector RAG. Building a PageIndex tree locally with PageIndex Flash now costs less than a vector index. LLMs only write the node summaries, using gpt-6-luna by default.

Sep 21, 2026

  • Cookbook. Runnable notebooks to start from — index a long PDF with PageIndex Flash, cite and highlight the source of an answer, retrieve document images, and estimate what a query costs. Open any of them in Colab.

Sep 16, 2026

  • New usage billing. PageIndex Cloud now uses simple usage billing — $0.01 to index a page, $0.001 / month to keep it active with unlimited retrieval.

  • Block-level citations. Cloud answers now cite the exact layout block rather than the page. Cloud documents indexed since September 13, 2026 carry blocks.

Aug 26, 2026

  • Introducing PageIndex Flash. A fast engine that builds the PageIndex tree for long, text-based PDFs. It runs on your own machine: the tree's structure is computed from the PDF layout without an LLM, and an LLM only writes the node summaries. Available now in the PageIndex SDK.