LlamaIndex releases OCRia-1, a 4B document-parsing VLM claiming SOTA on ParseBench at 87.4
llama_index · x · 2026-10-22
LlamaIndex launched OCRia-1, a frontier VLM for document parsing claiming new SOTA in accuracy, cost, and latency. It scores 87.4 on ParseBench and leads three third-party benchmarks: olmOCR-bench (83.5 overall, 92.9 on tables), PulseBench-Tab (84.1), and fr-bench-pdf2md (72.1). The 4B model specializes in complex tables (nested headers, merged cells, scanned grids) and unlabeled charts, outperforming same-class models like LightOnOCR-3, Chandra-OCR 2, and Mistral OCR 4.1 while running extremely fast.
More from Models
- Kimi K3.1 vs Qwen 4: both build a 3D helicopter with exploded views in ~15 minutes — Scobleizer · 2026-10-23
- FreshStack retrieval leaderboard adds Cohere Embed v5, EmbeddingGemma2, Perplexity late-interaction models — beirmug · 2026-10-23
- Together AI and Stanford's Next-Concept Modeling: 10-31x faster than diffusion LMs at equal quality — jiqizhixin · 2026-10-23
- Full-duplex competition will hinge on interaction timing, not just answer quality — kimmonismus · 2026-10-23
- China's Telos, seed-backed by HongShan and Hillhouse, builds omni-modal full-duplex models — kimmonismus · 2026-10-23
- Telos co-founder Bokai Xu co-designed MiniCPM-o's duplex architecture, bio shows — kimmonismus · 2026-10-23