Opus 5.5 is the best frontier model for parsing tables in PDFs.

We ran it through ParseBench, and it scored 93.9%, a 7%+ increase over Opus 5, and beating Fable/Gemini/Astra.

It's comparable to LlamaParse Agentic Plus in table quality, though it still struggles on charts, formatting, and layout. It's also 5.8c a page, which is too expensive to be a production OCR solution.

Full results on ParseBench: parsebench.ai

If you want Opus 5.5-level table quality, consistent parse quality + metadata for all edge cases, at the fraction of the cost, come check out LlamaParse. We have volume discounts for all modes and can help you set up customized routing between Agentic Plus, Agentic, and cost-effective: cloud.llamaindex.ai