ParseBench puts Claude Opus 5 near Opus 4.8 on docs, but cheaper rivals win on tables
llama_index · x · 2026-07-25
- A ParseBench evaluation found Claude Opus 5 is roughly on par with Opus 4.8 for document understanding.
- It does a bit worse on dense tables, but slightly better on charts and visual grounding.
- Gemini 3.6 Flash performs better on tables and is about half the price; LlamaParse agentic is stronger across the board, including tables, at roughly one-sixth the price.
- The takeaway: Opus 5 is fine for coding and knowledge work, but at about 8¢ per page and only average OCR performance, it is not a good choice for document parsing at scale.
More from Infra
- Huawei’s 950 NPU reportedly uses 4 dies and separate SerDes dies — teortaxesTex · 2026-07-25
- Databricks says Genie Code beat three coding agents on 401 real tasks at $0.55 each — DbrxMosaicAI · 2026-07-25
- SGLang v0.5.16 adds DSpark, cuts GLM-5.2 KV memory and supports Inkling on day one — BanghuaZ · 2026-07-25
- Alphabet’s future spending commitments jump to $811B as Stripe reportedly weighs OpenRouter — 创业邦 · 2026-07-25
- Nvidia faces a far more crowded AI-chip field in the West — teortaxesTex · 2026-07-25
- Fluidstack hosts San Francisco dinner on the future of AI infrastructure — MxMnr · 2026-07-25