ย
Never let a signup go cold again.
Turn emails into revenue with Brew. No credit card, free credits to try.
Turn emails into revenue with Brew. No credit card, free credits to try.
llamaindex.ai ยท newsletter
Explore this email design and adapt it to your own brand. Review the copy, links, and offer before sending.
Email Newsletter 1200x360 2026-08-12
Hi there, LlamaParser!
It's been a big week for document extraction: we released our latest research at ExtractBench.ai, the most comprehensive benchmark for information extraction on complex docs. Everything is full reproducible, we published the leaderboard, open sourced the eval harness, published the dataset as well as shared our paper on the methodology (all easily accessible through our blog below). Meanwhile, Meta and NVIDIA open-sourced models built to run local on your laptop. Grab a coffee, here's the rundown.
Discover LlamaParse โ most accurate document ingestion API for your agents.
GET API KEY
GET IN TOUCH
๐ข Product Updates
ExtractBench โ 4.8k real-world page from 370 enterprise documents, 14 extraction systems benchmarked head-to-head, and our new Agentic Plus Extract tier, posting 95.6% value F1 at 8.1ยข/page while commercial VLMs drop below 35% recall past 50 pages. Read more
Extract Agentic Plus โ the new top-tier extraction pipeline on the LlamaParse platform, built for the hardest documents: dense financial reports, scientific papers, and multi-column layouts, with chart parsing on by default. Holds 94.4% recall past 50 pages while commercial VLMs fall below 35%, at 8.1ยข/page. Read more
๐คฉ Community Updates
Webinar: Inside ExtractBench โ we'll dig into the numbers behind designing the benchmark and where a typical extraction pipeline is leaving accuracy, and sometimes cost, on the table: Aug. 26th โ> Save your spot
SF Billiards Night โ LlamaIndex, Judgement Labs, Modal, and ClickHouse are co-hosting pool night for AI founders and engineers to talk agents, infra, and production war stories. Thursday, Aug. 20 in San Francisco. RSVP
๐ AI News
Meta open-sourced Muse Glimmer, a 30B model built to run local agents on a single consumer GPU โ Apache 2.0, no cloud round-trip. Handy if you've ever wanted agents without also building a cloud bill. Read more
NVIDIA's local-AI roundup includes Nemotron 3.5 Lightning (4x faster token generation on consumer hardware) and NeMo Switchyard, an open-source router that sends each agent step to whichever model is cheapest for the job, cutting completion costs to about a third. We see you, routing is the whole game now. Read more
Grok 4.6 landed and is already claiming third place on Artificial Analysis, ahead of Kimi K3 and tied with GPT-5.6 Sol โ xAI says 4.7 is a month out. Fast-follow energy. Read more
xAI shipped Grok Bot, an always-on cloud agent that watches how you do a task once and then just does it, coordinating with other bots without you in the loop. Early beta Read more
Anthropic started watermarking every Claude output, everywhere, not just in the EU โ an imperceptible text mark plus C2PA-signed image metadata, in step with the EU AI Act's Art. 50 rules that took effect. Read more
Gradient_Newsletter
TRY FOR FREE
SEE PRICING
Building a new agent? Get in touch with our team
Need support? Join our Discord
Check out our GitHub for the latest code releases and documentation.
Thanks for reading this week's newsletter!
We are trying a new format and would love your feedback. Did we miss anything? Feel free to reply directly or email news@llamaindex.ai with your input.
LlamaIndex-logo
LlamaIndex, 405 Howard Street, San Francisco, California 94105, United States
Unsubscribe
Manage preferences
github
discord
X
YouTube
Outline_Text_Newsletter
