Unsiloed AI – #1 on olmOCR-Bench
Most of the document parsers fail on real world challenges like complex tables, handwritten documents, historical document scans, equations, multi-column layouts, complex reading order, etc. We built Unsiloed Parser to handle exactly these cases. Our latest parser v3.1 achieved #1 rank and scored 88.0 strict pass-rate on olmOCR-Bench. We ran the evaluation across 1,403 PDFs and 8,413 unit tests using the unmodified upstream Allen AI scorer (olmocr==0.4.27) and found Unsiloed beats 18 other OCR services, including GPT-5.5, Claude Opus 4.7, LlamaParse, Reducto, Azure Document Intelligence, AWS…
What it does
In the maker’s words, at launch
Most of the document parsers fail on real world challenges like complex tables, handwritten documents, historical document scans, equations, multi-column layouts, complex reading order, etc. We built Unsiloed Parser to handle exactly these cases. Our latest parser v3.1 achieved #1 rank and scored 88.0 strict pass-rate on olmOCR-Bench. We ran the evaluation across 1,403 PDFs and 8,413 unit tests using the unmodified upstream Allen AI scorer (olmocr==0.4.27) and found Unsiloed beats 18 other OCR services, including GPT-5.5, Claude Opus 4.7, LlamaParse, Reducto, Azure Document Intelligence, AWS Textract, and Unstructured. When we dug deeper into the failure cases, we found many errors were not OCR errors but things like \frac vs \dfrac, whitespace differences, or equivalent LaTeX renderings. We ran a secondary LLM-as-Judge evaluation to classify real misses vs semantic equivalents, which lifts the corrected score to 94.8 (explained deeply in the blog post). Blog with full methodology and examples: https://www.unsiloed.ai/blog/unsiloed-ai-achieves-1-rank-on-... Evaluation Code for reproducibility: https://github.com/Unsiloed-AI/unsiloed-olmocr-benchmark Feel free to post your messiest PDFs in the comment and we'll run it through Unsiloed parser and share the output here.
Does the same job
all alternatives →
- DADoctly AI – Accurate AI-Powered PDF to Markdown Parser2024 · ▲7
I’m one of the co-founders of Doctly AI. I wanted to share our story. We didn’t originally set out to build a PDF-to-Markdown parser. It all started when we were building a RAG solution for a company that deals with regulatory agencies. All of their data was in PDFs, and as it is apparently with lawyers, they like to print and scan documents to make it hard on their counterparts. These documents contained complex tables that barely make sense, are rotated, and handwriting is mixed in between. Many pages are number ruled and potentially rotated. We spent a lot of time trying to get clean data…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, May 2026
the whole month →
Parallel agents, diff reviewer, and multi-model comparisons
Dev tools · May 2026 · kilo.ai
- NW
Hey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok/s prefill and 1200 tok/s decode on consumer devices. We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention…
Life & fun · May 2026 · github.com
- FM
Dev tools · May 2026 · github.com