@llama_index
Anthropic says Opus 4.7 hits 80.6% on Document Reasoning β up from 57.1%. But "reasoning about documents" β "parsing documents for agents." We ran it on ParseBench. β Charts: 13.5% β 55.8% (+42.3) β huge β Formatting: 64.2% β 69.4% (+5.2) β Content: 89.7% β 90.3% (+0.6) β Tables: 86.5% β 87.2% (+0.7) β Layout: 16.5% β 14.0% (-2.5) β regressed Real chart gains, but at ~1.5Β’/page. Enterprise scale? Not yet. LlamaParse Agentic: 84.9% overall. ~1.2Β’/page. The frontier for general document understanding is long. No single model solves it. β https://t.co/h7SpuTWYVn