Hi there, LlamaParser!
It's been a busy June at LlamaIndex. Event season in San Francisco was in full swing, you may have even bumped into one of our billboards, but that didn't stop our engineering teams from shipping at full speed. There's a lot to catch up on, so without further ado...
Product Updates
- Announcing Retrieval Harness: LlamaParse now hands agents filesystem-level tools — list, grep, read, and hybrid retrieve — so they can traverse and verify a corpus instead of hoping one semantic search grabbed the right chunk.
- Markdown comes to LiteParse: LiteParse v2.1 is the fastest open-source, local PDF-to-markdown pipeline (~3ms/page) and tops all three parsing benchmarks we tested.
- LlamaParse MCP, restructured: one general endpoint at mcp.llamaindex.ai/mcp, plus per-product endpoints for Parse, Extract, Classify, Split, and Index. This gives your agent a narrower, more reliable tool surface and unlocks parallel workflows.
- Cost Optimizer, sharper: New improvements to LlamaParse's intelligent tier routing now more reliably ensures you always strike the right balance between cost and accuracy when processing large documents.
- Usage Tags (Enterprise): pass a Usage-Tags header on parse/extract, slice billing by team, environment, or any dimension. Up to 4 tags per request.
- User Metadata on parse jobs: Tag your requests, and your spend becomes filterable by those tags in the metrics API. Break down costs by customer, by team, by product line, whatever makes sense for your org.
Community Updates
- From messy PDFs to verifiable answers, with LiteParse and LanceDBA hands-on build pairing LiteParse for parsing with LanceDB for retrieval — the recipe for answers with citations you can actually trust.
- LlamaParse Platform is now a verified n8n community nodeIf you live in n8n, this is the shortest path from "PDF lands in a folder" to "structured JSON in your pipeline." No glue code necessary.
- Watch our founder and CEO speak at a full-house at Databricks Data+AI Summit (free to watch after sign up): Catch Jerry's Session Replay
- Grep vs. Embeddings with George He, Head of Engineering — the retrieval harness we built for enterprise agents, and when keyword search beats vectors: Watch Webinar
- Jerry at AI Engineer World’s Fair — our founder's take on where agentic document processing goes next, presented last week: Access the deck
AI News
- Anthropic extends access for Fable 5 until July 15th — the global redeploy still maintains a weekly usage limit. Join the chat
- Applications to attend OpenAI's DevDay 2026 are open: Apply by 7/10
- Best use of Fable 5 yet? Statistical analysis of World Cup odds Read now
- Where does Fable 5 rank on ParseBench.ai? We benchmarked Anthropic's latest model, check out the results below.
