Content History
Agentic spreadsheet extraction ๐ is hard to execute well from a price/performance standpoint because your agent needs to ingest a massive quantity of input tokens from a highly complex interface (e.g. the Excel file format), and emit a massive number of output tokens that accurate represents the source document. We've tuned an agentic harness within LlamaParse that can extract from even the most complex spreadsheets and schemas, and emit a large volume of structured output with extremely high accuracy. It can be used for any document, and now it can be used for spreadsheets too! Simply toggle "spreadsheet extraction" in the advanced processing options (see screenshots below) Come check it out: https://cloud.llamaindex.ai/
We've introduced native, agentic spreadsheet extraction ๐ into LlamaParse. Spreadsheets are a wildly different format from PDFs (or any other document format). They can span arbitrarily many rows/columns, lack any guarantees on tabular structure, and can have information
RTโLogan Markewich Obligatory plug for static embeddings. Are you using static embeddings in your projects? Would love to hear how people are leveraging them, whether for retrieval and beyond.
Static embeddings offer unmatched throughput, but they also suffer from accuracy loss compared to traditional embedding models. Can we make them better for retrieval? We tried several things: โ ๏ธ raw maxsim scoring on the per-token embeddings โ ๏ธ training a small adapter
static embeddings are extremely fast/cheap compared to dense embeddings, and still quite powerful. one issue is that they lose accuracy over longer text, since the tokens get pooled and averaged late interaction models score with MaxSim instead of average pooling, and avoid the pooling accuracy degradation. Can we apply these concepts to static models? This is a fun writeup by @LoganMarkewich - come check it out! https://www.llamaindex.ai/blog/exploring-static-embedding-retrieval
Static embeddings offer unmatched throughput, but they also suffer from accuracy loss compared to traditional embedding models. Can we make them better for retrieval? We tried several things: โ ๏ธ raw maxsim scoring on the per-token embeddings โ ๏ธ training a small adapter
RTโLlamaIndex ๐ฆ Static embeddings offer unmatched throughput, but they also suffer from accuracy loss compared to traditional embedding models. Can we make them better for retrieval? We tried several things: โ ๏ธ raw maxsim scoring on the per-token embeddings โ ๏ธ training a small adapter model โ ๏ธ changing the distillation training target and teacher While none of these saw the results we wanted, it's an excellent dive into static embeddings and how they do (and don't) work! Blog: https://www.llamaindex.ai/blog/exploring-static-embedding-retrieval
PDF parsing is fun because there's an infinite variety of enteprise documents ๐. For each document category, there's a long tail of work to build more precise bounding boxes, confidence scores, and domain-specific annotations so that you provide any downstream agent rich metadata without it having to reinvent this from scratch. Take forms for example. Besides simply outputting it into markdown, we put in the work to detect every annotation, field, checkbox, and section. That way you immediately get structured information as to whether a form is filled without a separate LLM extraction step. You also get source citations for free! Doing this well is hard. โ There's a rabbit-hole of optimizations you can do, including extracting annotations for every document type. โ It's hard to properly render visual formats like charts, handwriting into digitalized information. The more you skip this step, the more work you're creating for any downstream agent. โ Precise bounding boxes are a necessity for precise citations on any type of document. You can aggressively tune the model + harness so that the accuracy/cost on any document subtype is much more competitive than the frontier models. Whether you're parsing forms (see the enriched forms option in "processing options") or any other type of doc, check out LlamaParse ! https://cloud.llamaindex.ai/
We've introduced native, agentic spreadsheet extraction ๐ into LlamaParse. Spreadsheets are a wildly different format from PDFs (or any other document format). They can span arbitrarily many rows/columns, lack any guarantees on tabular structure, and can have information linked across multiple sheets. The native ways to deal with spreadsheets is usually through a code interpeter as opposed to OCR. We've introduced a tuned agentic engine (model+harness) that is equipped for large-scale schema guided extraction from spreadsheets. You can extract out dense sheets like balance sheets into clean structured fields. Check out the docs for how to integrate into the API: https://developers.llamaindex.ai/llamaparse/extract/guides/configuring-extract/#spreadsheet-mode On the UI, select "agentic plus", and see "spreadsheet options" in the advanced options section. Signup for LlamaParse here! http://cloud.llamaindex.ai/
Most extraction tools treat spreadsheets like PDFs. They flatten the file into text or markdown, then ask a model to infer the original structure. But spreadsheets depend on structure. Headers, formulas, merged cells, and hidden rows give every value its context. Strip that
RTโLlamaIndex ๐ฆ Most extraction tools treat spreadsheets like PDFs. They flatten the file into text or markdown, then ask a model to infer the original structure. But spreadsheets depend on structure. Headers, formulas, merged cells, and hidden rows give every value its context. Strip that away, and you map the right number to the wrong metric or period. That's why we built native spreadsheet extraction into the LlamaParse platform. Instead of flattening your workbook to text, it reads the raw cells directly and maps the data to your schema. Available today in beta on the agentic_plus tier. Give it a spin on your messiest .xlsx, .xls, or .csv files. Docs: https://developers.llamaindex.ai/llamaparse/extract/guides/configuring-extract/#spreadsheet-mode
We benchmarked 20+ open-weight models on easy-to-hard document extraction tasks through ExtractBench. The results are all available on @huggingface ๐ค ExtractBench is a schema-guided extraction benchmark that contains 4.8k+ pages across 8 domains and 67 document types, with a mix of short, medium, long docs and simple/complex schema.s The results reported is a measure of โvalue accuracyโ through unified F1. โ Qwen 3.8 leads the pack โ earlier generations of Qwen models are also quite strong โ kimi-k3 is the next best. GLM-5.3-flash and qwen 3.8 flash also just released today - hopefully will have results on these soon! Note : these results donโt include visual grounding (whether each value is mapped to the right bounding box) and confidence scores. When you include these, it is increasingly clear why specialized OCR tools (like LlamaParse) matter, since this is metadata thatโs hard to DIY by prompting the raw model. Come check out our HF leaderboard: https://huggingface.co/datasets/llamaindex/ExtractBench?leaderboard_base_model=false Learn more about ExtractBench here: https://www.extractbench.ai/
RTโsimon Computer needs document parsing
Parsing runs entirely on-device, so sensitive documents never leave the machine. On ParseBench-100, Computer scores 65.1% vs 34.6% for Hermes and 13.9% for Pi, in least time with fewest tokens.
If you are looking to parse large volumes of PDFs (and other document formats) while maximizing accuracy and cost, we're offering $1k in credits if you signup to the Pro plan! We're guaranteeing high quality and low cost. If you run into issues on the Pro plan we're always happy to open a Slack channel to help out. https://www.llamaindex.ai/pricing
๐ฆโ๏ธ Itโs a hot Llama summer, and it pays to be a PRO at document parsing. For a limited time, upgrade to the LlamaParse Pro plan and we'll triple your credits: 400K โ 1.2M credits in your first month. That's $1,000 (~80k pages) of parsing, free. No code, applied
How do startups differentiate and win against the frontier labs? - product UI/UX - brand - harness - data - workflow We'll talk about this with @GuangyuRobert over good food and good vibes ๐
Hosting the 2nd founder dinner in sf on August 31 with @GuangyuRobert at @Fundamental - the team behind @tryshortcutai Agents are increasingly capable of solving tasks for longer task horizons without human in the loop. This has rewarded "horizontal" products owned by the
View quoted postRTโCoreWeave Another special guest is hitting the stage at #FullyConnected26! @jerryjliu0 built @llama_index, the framework thousands of teams use to connect LLMs to their data. Come see Jerry and many more Sept 29th to Oct 1st in SF. Grab your spot here: https://www.utm.io/usfdT
We built one of the most comprehensive benchmarks for document extraction, and evaluated it across a lot of different systems: โ one-shot frontier VLMs โ frontier VLMs + coding agent harnesses โ one-shot open weight VLMs โ other document extraction tools (including LlamaParse) Document extraction is an extremely diverse task that covers many different types of docs. From simple schemas over short docs (e.g. resume extraction) to complex extraction docs (credit agreements, data room bundles). @disiok is leading this webinar. Come check it out! https://watch.getcontrast.io/register/llamaindex-inside-extractbench-benchmarking-document-extraction-for-agents
Every extraction API demos well on a clean invoice. But what about the scanned form, the nested table, the 40-page financial report with merged headers? We tested 14 frontier systems to find out. ExtractBench evaluates schema-guided extraction across 370 enterprise documents, 67
This summer, we brought in two first-year Waterloo interns (โrising sophomoresโ in ๐บ๐ธ) and tasked them with shipping complex production features that help push the frontier of our document processing and extraction capabilities. Yash created a dashboard that allowed customers to better understand the distribution of documents flowing through the system - and gives us anonymized metrics that we can use to improve our own capabilities. Jishnu helped to do deep research into grounded confidence scores for all of our document extraction capabilities. These scores are a lot more calibrated to real uncertainty about any value being extracted compared to our earlier implementation, allowing customers to trigger HITL in the right places within their document workflow. More on this coming soon ๐ Outside of work, I have fond memories of watching the Knicks final game 4 with them (the one where the Knicks came back down 29 points). Weโve historically brought on full-time engineers with a bit more experience, but Yash and Jishnu ramped up extremely quickly and exemplified the high-agency, high-energy scrappiness that we look for in anyone joining the @llama_index team. They werenโt old enough to drink but old enough to ship production code ๐. Weโre so happy they spent time with us so early in their careers, and wish them the best with the next 4 years of school. If youโre young but cracked, weโd love to chat. Weโre hiring across *a lot* of roles across product/engineering/research (as well as growth and GTM). Careers page: https://www.llamaindex.ai/careers
RTโLlamaIndex ๐ฆ Our 2nd founder dinner in SF co-hosted by @jerryjliu0 and @GuangyuRobert at @Fundamental - the team behind @tryshortcutai Talking about existing moats in the AI era. Frontier labs are moving past model APIs into vertical agents - ChatGPT Health, Claude for Legal. So where does the moat sit now? โ Agent engineering โ Infra optimization โ Domain evals and data โ Workflow expertise โ GTM and brand If you're a founder or CTO shipping agents in production and want to know what other teams are doing to maintain their moat. Request a seat ๐๏ธ https://luma.com/llamai-8hry
RTโTrue Ventures Coming to Connected Stack in SF โ @jerryjliu0, CEO & Co-founder of @llama_index, nominated by @GreylockVC. LlamaIndex is building document infrastructure for AI agents, with LlamaParse purpose-built for the complex documents that general-purpose models struggle with. See Jerry Liu on September 10 in San Francisco. Request an invite at http://connectedstack.ai.
One of the interesting properties weโve observed around schema-guided, complex document extraction tasks ๐ is that coding agent harnesses are good baselines (in terms of cost/accuracy) for longer documents. We tested Claude Code and Codex, along with specialized OCR tools (including LlamaParse) and raw VLMs. * On short documents, specialized OCR tools are generally a fraction of the cost of coding agents, with equivalent or higher accuracy * On longer documents, coding agents are a bit closer to the cost/accuracy Pareto curve (see bottom graph) Itโs an interesting result, though ultimately not surprising. Complex document extraction is a specialized reasoning task, and coding agents are effectively generalized reasoning harnesses. Over long documents, coding agents have more room to use a variety of tools to search snippets of the document instead of loading the entire document into context. They can also make use of prompt caching to reduce total token cost even as it expands multi-step reasoning. On the flip side, they do generate a baseline degree of token usage that proves to be wasteful for shorter docs compared to specialized extractors. This specific graph is in our Appendix D in the ParseBench paper, come check it out! ArXiv: https://arxiv.org/pdf/2607.29677 ExtractBench: https://www.extractbench.ai/
RTโLlamaIndex ๐ฆ Our CEO @jerryjliu0 is speaking at @CoreWeave Fully Connected 2026 in SF. ๐ค "Automating Document Work with Long-Horizon AI Agents" Sept 30, Moscone Center. Come see how long-horizon agents are taking on real document workflows in production. See you there! Checkout the full lineup ๐ https://www.coreweave.com/fully-connected-2026
we're offering exclusive startup credits and discounts to Tampa Bay Wave startups for pdf parsing come sign up here ๐ https://www.llamaindex.ai/startups
TIME Magazineโs list of โbestโ accelerators is your regular reminder that ALL media rankings are bullshit Their formula: โข 40% alumni recommendations โข 45% alumni category ratings โข 10% track record โข 5% expert recommendations Tampa Bay Wave beats YC ๐
We tuned an AI agent that can do large-scale document extraction from long docs (50+ pages, some with 10k-100k fields) with 94%+ accuracy ๐ It uses a harness + model set that is tuned specifically for reasoning over extracting out complex information from complex docs. Each extracted field comes with a confidence score as well as a bounding box denoting where it came from. It does 10-20% better in accuracy than generalized coding agent harnesses (e.g. Claude Code Opus 4.8 and Codex GPT-5.6). Check out the video below for a demo. The mode is called LlamaExtract Agentic Plus. Learn more about our extraction benchmark and agentic plus mode here: https://www.llamaindex.ai/blog/introducing-extractbench If you have very complex document extraction needs, come check out LlamaParse! https://cloud.llamaindex.ai/
Our "agentic plus" extractor in LlamaParse is great for extracting out massive volumes of fields (e.g. 10k-100k+ fields) from long documents (100-500 pages) We tested with a doc that contains a giant matrix of all creditors for FTX ๐ธ (75k fields, 114 pages). See screenshot
Our "agentic plus" extractor in LlamaParse is great for extracting out massive volumes of fields (e.g. 10k-100k+ fields) from long documents (100-500 pages) We tested with a doc that contains a giant matrix of all creditors for FTX ๐ธ (75k fields, 114 pages). See screenshot below. You can find full benchmark results on ExtractBench (http://extractbench.ai). In the meantime, come try out the mode in our playground! https://cloud.llamaindex.ai/
Introducing ExtractBench, the most comprehensive benchmark for information extraction from complex enterprise documents. The latest models are pushing the frontier of coding and knowledge work, but surprisingly they still struggle on complex doc extraction tasks in production. A
View quoted postWe think a lot about the distribution of all real-world documents, particularly for real-world use cases. We think about the distribution in terms of page length, field length, domain, and task type. Some documents have a single page but can have anywhere from 1-10 fields to 500-1000 fields to be extracted. Other documents are hundreds of pages long and can have 10k-100k fields. On long documents (e.g. a bankruptcy creditor matrix), commercial VLMs lose attention and create large swaths of missing or hallucinated values. The number of fields to extract per page also depends on the extraction task. 1. Needle-in-a-haystack: You find very specific information within a general document corpus. Generally 1-10 fields per page 2. Dense documents: typically extract from filled forms. Quite dense per page, though not super long in terms of total # pages. 3. Long-lists: Documents with both massive #'s of values per page (e.g. 100-1k fields per page) and massive numbers of pages. We'd love for you to explore our overall data taxonomy within ExtractBench, our comprehensive benchmark for real-world document extraction tasks. I love the diagram we show below in Figure 6, that offers one slice of the diverse set of data that we curate! Come check out the ArXiv paper: https://arxiv.org/pdf/2607.29677 ExtractBench: https://extractbench.ai/
Introducing ExtractBench, the most comprehensive benchmark for information extraction from complex enterprise documents. The latest models are pushing the frontier of coding and knowledge work, but surprisingly they still struggle on complex doc extraction tasks in production. A
View quoted postThere's a lot of real-world documents that are scanned, rotated, handwritten, or some combination of any of these elements. This week we created a comprehensive document extraction benchmark that contains documents tagged with various "perception challenges", along with other tags denoting task challenges, table structure, business domain. These docs include regulatory filings, hand-filled tax forms, photocopied docs, sensor noise, and more. Codex is surprisingly good at scans, but not great on rotated docs. OCR solutions are reasonable on rotations/handwriting but struggle on more general scans. Check out ExtractBench! ArXiv: https://arxiv.org/pdf/2607.29677 Site: https://www.extractbench.ai/
Every document extraction system has a perception blind spot. We mapped them. For ExtractBench, we tested 14 systems on documents that weren't born digital: 1950s regulatory filings, hand-filled tax forms, and pages degraded with fax thresholding, photocopier tone curves, sensor
View quoted post
Our UK reps didnโt remember to smile ๐ We recently opened a @llama_index office in London ๐ฌ๐ง. It is three people strong (Frank, Max, Joe), with two more people starting soon. It also has a beautiful view of the Thames. Weโre hiring people across SF, London, and (soon) NYC in a ton of different functions: * Member of Technical Staff: across core product engineering, infra, and AI research * Forward deployed AI engineers * Mid-market and enterprise account executives BDRs * Field marketing and devrel * GTM and technical recruiting Weโre growing quickly - weโve doubled in size in the past few months - and would love to bring more folks to join our journey ๐ Come check out our careers page: https://www.llamaindex.ai/careers
RTโLlamaIndex ๐ฆ ๐ฆโ๏ธ Itโs a hot Llama summer, and it pays to be a PRO at document parsing. For a limited time, upgrade to the LlamaParse Pro plan and we'll triple your credits: 400K โ 1.2M credits in your first month. That's $1,000 (~80k pages) of parsing, free. No code, applied automatically. Upgrade today ๐๏ธ http://llamaindex.ai/pricing
This week we launched the world's most accurate document extraction agent over real-world documents. Introducing LlamaExtract Agentic Plus ๐ซ . It is a complete document extraction model+harness engine that can convert even the most complex document types/lengths/schemas into clean, grounded, structured output. We benchmarked it over ExtractBench - a comprehensive set of real-world enterprise documents. It is measured over various dimensions: document length (short/medium/long), task (long-list, needle in haystack, dense docs), perception (rotated, scanned, handwriting), table structured (merged headers, cross-page tables, table within cells), and business domains. We compared it against a variety of coding agent harness to document extraction solutions. LlamaExtract Agentic Plus is not only highly accurate, it is quite price efficient relative to its capabilities (25-50% the price of Claude Code/Codex). You can see the full release of ExtractBench here: https://www.llamaindex.ai/blog/introducing-extractbench If you want to check it out, come sign up for an account on LlamaParse: https://cloud.llamaindex.ai/
Introducing ExtractBench, the most comprehensive benchmark for information extraction from complex enterprise documents. The latest models are pushing the frontier of coding and knowledge work, but surprisingly they still struggle on complex doc extraction tasks in production. A
View quoted postExtractBench is one of the most comprehensive benchmarks for real-world document extraction. โ It covers 4869 pages, across 67 document types, spanning 8 real-world domains: finance, energy, gov, auto, supply chain, healthcare, legal, real estate โ It covers a distribution of short, medium, and long documents โ It covers a variety of very complex table edge cases: tables with over 1k rows, nested tables within cells, cross-page tables, and more โ It covers scans, handwriting, and rotated pages We benchmarked across 14 different VLMs, coding agents, and document extraction APIs. Everything is fully public on our blog, ArXiv, Github, and HuggingFace: Blog: https://www.llamaindex.ai/blog/introducing-extractbench Hugging Face: https://huggingface.co/datasets/llamaindex/ExtractBench Github: https://github.com/run-llama/ExtractBench ArXiv: https://arxiv.org/pdf/2607.29677
Introducing ExtractBench, the most comprehensive benchmark for information extraction from complex enterprise documents. The latest models are pushing the frontier of coding and knowledge work, but surprisingly they still struggle on complex doc extraction tasks in production. A
View quoted postA downside with VLM-based parsing is that theyโre generally slower than text-based heuristic approaches. As a result they add latency to any ad-hoc file processing *in-the agent loop* (e.g. if you upload a file to Claude) We built LiteParse to be a really good text-based, heuristic extractor so that you can use it as the default parser in the agent loop. It takes 4ms to parse 200 pages โก๏ธ, is more accurate than other OSS parsers, and you can install it within Claude Cowork/Code or Codex in one-line with our agent skill. It also supports 50+ different doc formats and has a native complexity router so you can route it to OCR/VLMs for any complex pages Come check it out! Repo: https://github.com/run-llama/liteparse
The best "raw" frontier model for document parsing is gemini 3 flash, but the issue is that since then the flash models have gotten 3x more expensive while flatlining on visual recognition across complex documents. This has been the case for every frontier model including the latest OpenAI/Anthropic models - see the diagram below for GPT (since then we've also benchmarked 5.6) In the meantime, hybrid approaches like LlamaParse that blend specialized VLMs with a text engine offer better performance; our own LlamaParse accuracy has increased 15% over tables and charts. If you have document OCR needs and are thinking about using a frontier model, you might as well come check out LlamaParse! https://cloud.llamaindex.ai/ We have a full eval harness through ParseBench that you can configure over your own docs: https://www.parsebench.ai/
Document OCR is not Getting Commoditized (by Frontier Models) The most common question I get is whether frontier models are going to eat all document processing solutions - just screenshot the page and feed it to your favorite frontier model. 1๏ธโฃ Frontier models are flatlining
My entire programming stack was tmux + vim for like 5 years before I begrudingly switched to VS code circa 2020
People shocked by this feature are just late to the party Itโs called โTmuxโ folks
LiteParse can now extract structured data from your PDF in milliseconds: โ checkbox states โ annotations โ vector graphics โ word-level bounding boxes It is the most comprehensive, accurate (and fast) free/open-source document processor out there. If you plug it into a coding agent to process digitalized PDFs, you will not only be able to give it the right context, but also give it grounding back to the source. For more complex docs that require VLM processing, you can always use the in-built router to route it to a VLM based solution, like LlamaParse. Come check it out: https://developers.llamaindex.ai/liteparse/guides/extraction/ LiteParse: https://github.com/run-llama/liteparse Complexity guide: https://developers.llamaindex.ai/liteparse/guides/complexity/
You shouldn't need a vision model to know your PDF has checkboxes. LiteParse can now pull structured data directly from your PDFs: form field values, checkbox states, annotations, embedded images, vector graphics, tagged document structure, and word-level bounding boxes, all
We've built a new feature in LlamaParse that lets you automatically extract any complex form into a structured JSON output ๐๐ค The best part is there's no schema needed! We will systematically detect and extract out every single form key and corresponding form value from the document (blank if not filled). Simply set `processing_options.forms='enrich'` Check it out: http://login.llamaindex.ai/sign-up API docs: https://developers.api.llamaindex.ai/api/resources/parsing/methods/get/#
Parsing a W-2 into markdown was always the easy part. Getting the fields out was a second pipeline: define a schema, map the fields, handle the edge cases. Set processing_options.forms='๐ฒ๐ป๐ฟ๐ถ๐ฐ๐ต', and LlamaParse returns a dedicated JSON with field names, values, and checkbox
RTโLlamaIndex ๐ฆ POV: It's the last day at @Ai4conferences and your: ๐งข colorful hats and stickers are all over the conference floor ๐จโ๐ซ CEO, @jerryjliu0 , delivered a speech to a full room ๐ฝ๏ธ executive dinner brought in some of the best and brightest in AI ๐ feet might hurt and new white shoes have a few scuffs, but still ๐ฅด
no shade but these slides look like a school project lmao
We created a pitch deck to tell a handful of VC firms about us and what we were up to (a fun experience!). Hereโs a few slides about our background and some of the things weโve worked on from the pitch deck (it was fun putting together the list of people in our teams who have
Document OCR is not Getting Commoditized (by Frontier Models) The most common question I get is whether frontier models are going to eat all document processing solutions - just screenshot the page and feed it to your favorite frontier model. 1๏ธโฃ Frontier models are flatlining in document understanding performance. Incremental releases in every model version (gpt 5.5 -> 5.6 sol, Gemini 3.5 flash -> 3.6 flash, opus 4.8 -> opus 5) are not improving visual understanding benchmarks. 2๏ธโฃ The Pareto frontier for document OCR is much higher than the frontier models, and will always remain much higher. Weโve carefully tuned our agentic and cost-effective modes to solve a long tail of complex edge cases (dense tables, charts) that frontier models donโt care about. Also for equivalent performance, thereโs always ways to get much lower cost. 3๏ธโฃ Even if they were getting better, you can distill / posttrain them into much more parameter efficient models for a fraction of the cost. Different document pages can be routed to different processors of varying complexity. 4๏ธโฃ Every startup is doing the same thing right now. Focusing on one task means you can always hillclimb that task more effectively than a general model over the task, in terms of accuracy/cost/latency. Check out the blog: https://www.llamaindex.ai/blog/document-ocr-is-not-getting-commoditized LlamaParse has gotten a LOT better in the past few months. Come check it out! https://cloud.llamaindex.ai/
"OCR is just a feature now. Frontier models will eat it." We hear this constantly. The data says otherwise. Across three GPT generations, parsing accuracy gained ~24 points, while cost per page 4x'd. And the newest frontier models still trail specialized parsers. Read more
this is the agi version of the paypal mafia
Announcing Discovery Loop! I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine
RTโLlamaIndex ๐ฆ Parsing a W-2 into markdown was always the easy part. Getting the fields out was a second pipeline: define a schema, map the fields, handle the edge cases. Set processing_options.forms='๐ฒ๐ป๐ฟ๐ถ๐ฐ๐ต', and LlamaParse returns a dedicated JSON with field names, values, and checkbox states alongside your standard markdown. No schema. No second API call. Now available in beta on all paid plans. Signup for 10k free credits: https://login.llamaindex.ai/sign-up Docs: https://developers.api.llamaindex.ai/api/resources/parsing/methods/get/#(resource
RTโClelia Bertelli (๐ฆ/acc) Yesterday was my last day at LlamaIndex๐ฆ It was an incredible journey. I had the opportunity to work alongside so many brilliant people, learn from them, and grow both as an engineer and as a person. Iโll always be grateful for this invaluable chapter :) The LlamaIndex team is truly awesome, and Iโm sure theyโll continue to build and ship great things. I wish everyone on the team all the best for the future! As for me, Iโll be sharing soon whatโs next๐ฉโ๐ Upwards and onwards๐
RTโLlamaIndex ๐ฆ ๐ฆ is coming coming to Las Vegas for @Ai4Conferences next week (Aug 4โ6). Three ways to find us: ๐ Booth 1561 on the Expo floor ๐ค @jerryjliu0 will be speaking about unlocking document context for AI agents ๐ฅ Waitlist for our (very oversubscribed) exec dinner at Tao Bistro (find us on Luma) Come say hi ๐
We created a document OCR router that can estimate the complexity of every single page and parse it with the relevant mode ๐ซ * Some pages are full of native text, which can be directly handled with LiteParse * Some pages contain scanned images or tables, which can be handled with our cost-effective or agentic tiers * Some pages contain garbled text - it decodes to garbage - so you need a heavier-weight VLM to interpret the text elements * Some pages contain heavy visuals like labeled/unlabeled charts or diagrams, which also require deeper visual processing You actually get the routing capabilities for free with the `is_complex` toggle in LiteParse. The next step is figuring out what the relevant VLM modes are to parse varying page complexities. That's well served by LlamaParse! Blog: https://www.llamaindex.ai/blog/parse-gateway-smart-page-level-document-parser-routing LiteParse: https://github.com/run-llama/liteparse LlamaParse: https://cloud.llamaindex.ai/
Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing a tradeoff between cost, speed, and accuracy across the whole document. We
OpenAIโs entire growth loop: new models, price cuts, and Tiboโs token resets
major price cuts today: *80% drop for GPT-5.6 Luna, now $0.20 per million input tokens and $1.20 per million output *20% drop for GPT-5.6 Terra, to $2/$12 *GPT-5.6 Sol gets Fast mode in the API, up to 2.5x the speed for 2x the price, same intelligence
We have an idea for dinner 2 already :) But what ideas are you all interested in? - technical topics like rl envs, continual learning, cloud agents, world models - general startup / company building from plg, growth, GTM, talent, etc
Yesterday I cohosted a dinner with @dexhorthy with a wonderful group of founders, to talk about agent loops and loop engineering. Some interesting insights: * Most of our group was *not* actively using /loop in Codex/Claude Code * You can build long-running autonomous agent
When an incumbent disrupts a startup
OH. MY. GOODNESS. CITADEL HAS BOUGHT A MAJORITY OF THE PUBLIC ASSETS FROM LEOPOLD'S SITUATIONAL AWARENESS FUND. So...Citadel scares everyone on Tuesday about a surprise rate hike during FOMC that WE ALL KNEW was not going to happen... On Wednesday, the entire market freaks
Introducing the Parse Gateway There's been an explosion of interest in model routing - you don't always need the best model for every task. That is especially true for document parsing ๐๐: - Some pages are text-heavy and relatively simple. These pages can be handled by cheap, fast text-based parsers - Other pages are more complex with a variety of dimensions: scans, dense tables, complex diagrams and charts, garbled binary text, and more. The trick is to use a good parse router that can optimize the parsing mode depending on page complexity! The Parse Gateway is based on the `is_complex` functionality we introduced within liteparse, which estimates the complexity of a page at the page level. Depending on the complexity, it routes the page to our various modes within LlamaParse: cost-effective, agentic, and agentic plus. The gateway code itself is open-source, but you can also directly find an implementation within our MCP server Check out the blog! https://www.llamaindex.ai/blog/parse-gateway-smart-page-level-document-parser-routing LlamaParse: https://cloud.llamaindex.ai/
Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing a tradeoff between cost, speed, and accuracy across the whole document. We
RTโClelia Bertelli (๐ฆ/acc) So great to see this one out! I really liked working on routing, and I feel like it's such a hyped topic for LLMs/agents, but very under-explored in other fields
Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing a tradeoff between cost, speed, and accuracy across the whole document. We
RTโLlamaIndex ๐ฆ Not every page in your PDF needs the same treatment. A scanned cover, a dense table, a clean text page, a figure-heavy diagram.... most parsing pipelines throw all of them at the same parser, forcing a tradeoff between cost, speed, and accuracy across the whole document. We built ๐ฃ๐ฎ๐ฟ๐๐ฒ ๐๐ฎ๐๐ฒ๐๐ฎ๐ to fix that. It uses LiteParse's ๐ช๐ด_๐ค๐ฐ๐ฎ๐ฑ๐ญ๐ฆ๐น functionality to estimate complexity page-by-page and routes each page to the right tier based on why it's complex (scanned, sparse text, garbled encoding, vector text, embedded images) and how severely. Simple pages get parsed for free, in-process. Genuinely hard pages get routed to more capable LlamaParse tiers. No manual heuristics, no flat one-size-fits-all pipeline. And it's also available as an MCP server, so your agents can estimate complexity and choose a parsing tier themselves! The result: you stop paying premium prices for pages that never needed it, without sacrificing accuracy on the ones that do. ๐๏ธ Full writeup: https://www.llamaindex.ai/blog/parse-gateway-smart-page-level-document-parser-routing ๐ฉโ๐ป Try it: https://github.com/run-llama/parse-gateway
Yesterday I cohosted a dinner with @dexhorthy with a wonderful group of founders, to talk about agent loops and loop engineering. Some interesting insights: * Most of our group was *not* actively using /loop in Codex/Claude Code * You can build long-running autonomous agent loops through multi-agent handoffs, event triggers, orโฆ.just stacks of cron jobs (?) * Almost everyone believes that no one will be reviewing code in 1-2 years. * The more interesting question is whether weโd be reviewing *anything* in 1-2 years. * AI is still a bit of a skill issue. Humans are responsible for maximizing AI output and reducing slopification. * Will human intellect provide alpha as models get better, or will the playing field be leveled? Most people think it will be leveled a bit, but there is a need for humans to provide alignment, guardrails, judgment, creativity. * The minimum amount of context you need for AI could be just the codebase with some documentation. Any research/plan files are for one-off tasks and not meant to be maintained. Having a self-organizing wiki is nice but adds complexity. If you missed this one, weโll be hosting more dinners like this on a regular cadence! If you have thoughts on what we should talk about, let us know :) (e.g. continual learning, RL envs, competitive differentiation vs. Anthropic, etc.)
