LlamaIndex releases OCRia-1, a 4B document parsing VLM topping ParseBench at 87.4
llama_index · x · 2026-10-27
LlamaIndex launched OCRia-1, a 4-billion-parameter vision-language model for document parsing, claiming SOTA on accuracy, cost, and latency. It scores 87.4 on ParseBench (best under 35B params, 12.1 points ahead), and leads table benchmarks including olmOCR-bench (92.9 on tables), PulseBench-Tab (84.1), and fr-bench-pdf2md. It specializes in complex tables (nested headers, merged cells, borderless grids, scans) and unlabeled charts, and is now powering LlamaParse. The team also published attention-map visualizations showing how attention shifts while predicting table cells.
Related event: LlamaIndex releases OCRia-1, a 4B document parsing model setting new SOTA(3 posts)→
More from Models
- Open-weights chart shows xAI at 0% despite Musk's 2024 open-source pledge, DeepSeek at 100% — kingslyj · 2026-10-27
- Baidu's open-source Unlimited OCR tops 10M HuggingFace downloads, 28.3K GitHub stars — xiaohu · 2026-10-27
- Unlimited OCR 2 ingests 3,600-page Harry Potter in one go, cutting KV cache from 47GB to 1.77GB — xiaohu · 2026-10-27
- Models spontaneously run internal expert debates when solving hard problems, nobody designed it — profjamesevans · 2026-10-27
- Speculation: xAI's announcement is likely Index, not Fable 5.5 — ChrisGPT · 2026-10-27
- AMD and vLLM team release open 27B multimodal decision model d3-ultra, #2 on Jev Decision Index — vllm_project · 2026-10-27