WHAT NEW IN X10.29 THU

X 上的 AI 合集档案 · 22 个合集 · 60,637 条

gpt-6IMAGE

GPT-6 Astra 文档抽取实测:短文档 97.2% 新 SOTA

ExtractBench 实测 Astra 短文档抽取达 97.2% 新 SOTA,但长文档仅 31.7%,每页成本约 11 美分。

Jerry Liu
@jerryjliu02026.09.14收录原推发布于 2026.09.05
GPT-6 Astra 文档抽取实测:短文档 97.2% 新 SOTA
Jerry Liu @jerryjliu0原推发布于 2026.09.05
在 X 上打开原推

We benchmarked GPT-6 Astra on hard document parsing and extraction tasks. It is the best frontier model for one-shotting document extraction over short and medium-sized documents: 📊 97.2% over short documents, achieving a new SOTA on our benchmark 📊 90.6% over medium documents But a few caveats: * it's pricy! The average price per page is 11c which is 10x our cost-effective document extraction solution * one-shot performance over long documents is inherently tricky, it only achieves 31.7% * the native OCR capabilities of Astra isn't that much better than other frontier models (e.g. Fable 5.1, gpt-5.6-sol). It does very well on table understanding, but isn't great on charts, layout, and semantic formatting. Full results for document extraction on ExtractBench: https://www.extractbench.ai/

x.com/jerryjliu0/status/2096258549131165746采集于 2026.09.14 15:53